Children tooth trauma intelligent grading and dynamic processing decision-making auxiliary system

Abstract

The invention relates to the technical field of artificial intelligence auxiliary medical treatment, and discloses a child tooth trauma intelligent grading and dynamic processing decision auxiliary system. A trauma identification module; an injury condition grading module; a processing suggestion generation module; a prognosis risk dynamic evaluation module; and a man-machine interaction module. The method comprises the following steps: acquiring an image containing child tooth trauma information and clinical data; utilizing a multi-label classification model to identify a trauma type; calculating a comprehensive injury condition score through a hierarchical multi-modal scoring network and determining an injury condition grade; calling a knowledge graph and a rule engine to generate and dynamically correct a processing suggestion; and continuously updating the prognosis risk assessment result based on the time sequence review data. The invention aims to solve the problems of strong evaluation subjectivity and complex decision in the diagnosis and treatment of the tooth trauma of children, provides standardized and intelligent decision assistance for clinic, and realizes closed-loop prognosis management.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence assisted medical technology, in particular to an intelligent grading and dynamic processing decision assistance system for children's dental trauma. BACKGROUND

[0002] Children's dental trauma is very common in clinical practice, which is often caused by accidental collision or falling. Timely and accurate treatment of such trauma is crucial, otherwise it may cause a series of serious long-term complications such as pulp necrosis, root resorption and even jaw development disorder.

[0003] In the prior art, the diagnosis and treatment decision of dental trauma largely depends on the personal clinical experience of the doctor on duty. This strong subjective diagnosis and treatment mode, combined with the uneven distribution of professional resources among different regions and medical institutions, leads to difficulty in unifying the diagnosis and treatment standards, and the timeliness and final effect of the treatment are also uneven. At the same time, although there are industry standards such as the International Dental Trauma Guidelines (IADT), these static text guidelines lack interactivity and cannot provide dynamic and individualized decision support for the specific conditions (such as age, complications, and injury combination) of each child.

[0004] Therefore, the present application proposes an intelligent grading and dynamic processing decision assistance system for children's dental trauma to solve the problems of the prior art. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides an intelligent grading and dynamic processing decision assistance system for children's dental trauma, which solves the problem that current diagnosis and treatment of children's dental trauma mainly relies on the subjective experience of doctors and lacks standardized and individualized decision support tools.

[0006] To achieve the above purpose, the present application realizes the following technical solutions: The present application provides an intelligent grading and dynamic processing decision assistance system for children's dental trauma, which comprises: A perception acquisition module for acquiring multi-modal data including tooth image; A trauma recognition module for analyzing the tooth image in the multi-modal data to identify the type of dental trauma and detect whether there is a concurrent occlusion relationship disorder, and combining the type of dental trauma and the detection result of the concurrent occlusion relationship disorder into a structured recognition result; An injury grading module for receiving the structured recognition result and combining other information in the multi-modal data, comprehensively scoring the injury through a hierarchical multi-modal scoring network, and outputting a standardized injury grade; a processing suggestion generation module configured to generate a structured processing suggestion based on the standardized injury grade and according to a detection result about a concurrent occlusal relationship disorder in the structured identification result; a human-computer interaction module connected with the perception acquisition module and the processing suggestion generation module respectively, configured to perform information input and processing result display of the user terminal, and specifically configured to receive the multi-modal data input by the user for acquisition by the perception acquisition module, and display the structured processing suggestion output by the processing suggestion generation module to the user terminal.

[0007] As a preferred scheme, the perception acquisition module locally deploys and runs an edge computing model on the mobile terminal. The edge computing model analyzes the camera view screen of the mobile terminal in real time to evaluate the image acquisition quality and acquisition angle of the current screen. Based on the evaluation result, the perception acquisition module outputs real-time interactive guidance instructions to the user terminal, which are used to guide the user to adjust the shooting behavior until the edge computing model determines that the tooth image meets the preset standardized acquisition requirements.

[0008] As a preferred scheme, the trauma identification module uses a multi-label classification model including a shared feature extraction backbone network and parallel output branches to analyze the tooth image. The module generates a damage area heat map representing the corresponding position of the tooth trauma type in the tooth image while outputting the structured identification result. The structured identification result includes a tooth trauma multi-label probability vector generated by the classification output branch and an occlusal disorder feature vector generated by the feature extraction output branch. The tooth trauma multi-label probability vector indicates the probability of the existence of each of the plurality of preset tooth trauma types; the occlusal disorder feature vector represents the occlusal state of the tooth in the form of numerical feature data, the value of which is associated with the clinical severity of occlusal disorder, and can specifically be a Boolean flag indicating whether there is occlusal disorder and a quantitative score indicating the severity of occlusal disorder.

[0009] As a preferred scheme, the injury grading module takes the structured identification result output by the trauma identification module as the core input, and combines the time-sensitive information and development information in the multi-modal data to calculate a comprehensive injury score. During this calculation process, the injury grading module is further configured to call preset differentiated risk assessment rules for different tooth types (primary teeth or permanent teeth) and development stages according to the tooth type and the age of the child in the development information, to calculate the time-sensitive and development risk scores. The comprehensive injury score is calculated according to the following formula: ; wherein, to synthesize the injury score; to synthesize the tooth structure damage score; to synthesize the tooth displacement damage score; to synthesize the occlusion impact score; to synthesize the timeliness and development risk score; corresponding to , , , respectively.

[0010] As a preferred solution, the treatment suggestion generation module comprises a knowledge graph and a rule engine. The knowledge graph is used to structure the standard dental trauma treatment guidelines and the clinical expert experience into a graph data structure containing medical concept nodes and semantic relationship edges. The rule engine runs based on the knowledge graph, and its execution process is: by traversing the medical concept nodes and semantic relationship edges in the graph data structure, reasoning out the standard treatment path matching the standardized injury level; when the detection result about concurrent occlusion relationship disorder in the structured recognition result is true, according to the quantitative score in the occlusion disorder feature vector, inserting additional complication treatment nodes or adjusting the treatment parameters of existing nodes in the standard treatment path to generate the final structured treatment suggestion.

[0011] As a preferred solution, the human-computer interaction module constructs a doctor-end professional interface and a parent-end guidance interface. The module adapts the structured treatment suggestion output by the treatment suggestion generation module to generate a detailed diagnosis and treatment plan for professional personnel and a graphic nursing guide for non-professional personnel respectively, and then displays the detailed diagnosis and treatment plan on the doctor-end professional interface and the graphic nursing guide on the parent-end guidance interface.

[0012] As a preferred solution, the system further comprises a prognosis risk dynamic assessment module. The module is connected with the human-computer interaction module and is used to receive the time-series review data uploaded by the user during follow-up, which includes current tooth image and clinical review parameters. The prognosis risk dynamic assessment module performs feature-level difference comparison between the current tooth image and the historical tooth image to quantify the risk indicators in the image dimension, and fuses the quantified risk indicators in the image dimension, the clinical review parameters, and the prognosis risk assessment result at the last time point to update and output the current dynamic prognosis risk assessment result.

[0013] Furthermore, the dynamic prognostic risk assessment module calculates the current dynamic prognostic risk assessment result using an autoregressive model. The current dynamic prognostic risk assessment result... The calculation formula is defined as follows: ; In the formula, This is the current dynamic prognostic risk assessment result; This is the prognostic risk assessment result from the previous time point; Risk value for image dimension; Risk values ​​for clinical parameters; This is the historical risk attenuation coefficient; The weighting coefficients for the risk values ​​of the image dimension; The weighting coefficients for the risk values ​​of clinical parameters; Basic risk constant; Among them, the image dimension risk value It is obtained by comparing the image of the affected tooth at the current follow-up visit with the image of the affected tooth at the previous time point using feature level difference. The core features of the comparison include at least one or more of the following: changes in crown color, changes in gingival morphology and color, and changes in periapical imaging.

[0014] Furthermore, the dynamic prognostic risk assessment module is also used to assess the current dynamic prognostic risk assessment results. When a preset warning threshold is exceeded, an active warning command is triggered, and the warning information is pushed to the user terminal through the human-computer interaction module.

[0015] A second aspect of this invention provides a method for intelligent grading and dynamic treatment decision support of dental trauma in children, applied to the aforementioned system, the method comprising the following steps: S1. Receive multimodal data, including images of the affected tooth, through the user terminal; S2. Analyze the image of the affected tooth in the multimodal data to identify the type of dental trauma and detect whether there is concurrent occlusal disorder, and combine the type of dental trauma and the detection results of concurrent occlusal disorder into a structured identification result; S3. Receive the structured recognition result and combine it with other information in the multimodal data, and use a hierarchical multimodal scoring network to comprehensively score the injury and output a standardized injury level. S4. Based on the standardized injury level and the detection results of concurrent occlusal disorder in the structured identification results, generate structured treatment suggestions; S5. Display the structured processing suggestions on the user terminal.

[0016] The application provides a child tooth trauma intelligent grading and dynamic processing decision assistance system. 1、The application ensures the standardization of input tooth image by setting interactive guidance instructions at the front end of collection, which provides guarantee for the consistency of subsequent analysis. More importantly, the trauma recognition module adopts a multi-label classification model containing a shared feature extraction backbone and a parallel output branch, which can not only analyze and efficiently recognize multiple tooth trauma types at one time, but also specially detects concurrent occlusion relationship disorder which is easily ignored in clinical practice, and generates a damage area heat map to provide visual evidence. This makes the diagnosis process less dependent on the personal experience of doctors, provides a more objective, complete and traceable identification result, and reduces the risk of misdiagnosis and missed diagnosis.

[0017] 2、The application establishes a multi-dimensional comprehensive injury scoring system, which overcomes the fuzziness and subjectivity of traditional grading methods. The system quantitatively fuses the direct indicators such as tooth structure damage and tooth displacement damage recognized by machine vision with indirect risk factors such as timeliness information represented by delayed time and development information represented by child age through a hierarchical multi-modal scoring network. The system can intelligently identify whether the tooth is a deciduous tooth or a permanent tooth according to the development information, and call relevant risk assessment rules to differentially weight the unique risks of different stages such as deciduous tooth root resorption or permanent tooth root development. Therefore, the standardized injury grade output by the system is no longer a simple qualitative description, but a precise quantitative result that comprehensively considers damage, timeliness, adaptive development risk and other multiple dimensions, providing a solid scientific basis for subsequent development of appropriate treatment strategies.

[0018] 3、The application converts the standardized clinical guidelines into a logic path executable by a machine by constructing a knowledge graph and a rule engine. The generation of treatment recommendations is not a simple matching of standard paths, but has a dynamic correction logic. The logic can intelligently adjust the initially generated standard treatment path according to the quantitative score of concurrent occlusion relationship disorder output by the trauma recognition module. The final structured treatment recommendation can fully consider the individual's complications while following evidence-based medicine, achieving highly individualized precision decision assistance.

[0019] 4、The application extends the diagnosis and treatment process from one-time evaluation to long-term follow-up management, and establishes a prognosis risk dynamic evaluation module. The module receives and processes the user uploaded time series review data continuously, uses the autoregressive model, compares the feature differences between the current and historical tooth images, and also integrates the parent feedback clinical review parameters. The prognosis risk evaluation results can be dynamically quantified and updated, potential recovery abnormalities can be found in time, and active warning instructions can be triggered when the risk value exceeds the threshold, realizing the whole cycle closed-loop management from diagnosis and treatment to continuous monitoring and active intervention.

[0020] 5、The application designs a doctor end professional interface and a parent end guidance interface, and differentiates the information adaptation of the same structured treatment suggestion according to the cognitive differences of different user groups. For doctors, the system presents detailed diagnosis and treatment schemes containing technical details to support professional decision-making; for children's parents, the system automatically generates a family care guide with pictures and text that is easy to understand. This double-channel information distribution mechanism effectively improves the understanding and execution compliance of parents to the care plan without increasing the workload of doctors, thereby promoting doctor-patient cooperation and providing protection for improving the overall treatment effect. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The system structure block diagram of the application; Figure 2 The intelligent grading and dynamic processing decision assistance method flowchart of child tooth trauma of the application; Figure 3 The trauma identification and visual positioning schematic diagram of the application; Figure 4 The processing suggestion generation logic schematic diagram of the application; Figure 5 The double interface schematic diagram of the man-machine interaction module of the application.

[0022] Among them, 10, a perception acquisition module; 20, a trauma identification module; 30, a trauma grading module; 40, a processing suggestion generation module; 50, a man-machine interaction module; 60, a prognosis risk dynamic evaluation module. DETAILED DESCRIPTION

[0023] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the application specification. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.

[0024] REFERENCE Figure 1 , Figure 1is a structural block diagram of a child dental trauma automatic grading and treatment advice generation system according to an embodiment of the present application. The system comprises a perception acquisition module 10, a trauma identification module 20, an injury grading module 30, a treatment advice generation module 40, a human-computer interaction module 50, and a prognosis risk dynamic assessment module 60.

[0025] The perception acquisition module 10 is used to acquire multi-modal input data related to child dental trauma. The human-computer interaction module 50 is in bidirectional communication connection with the perception acquisition module 10, and is used to provide an interface for data input. The output end of the perception acquisition module 10 is connected to the input end of the trauma identification module 20.

[0026] The trauma identification module 20 is used to analyze the standardized data output by the perception acquisition module 10, and identify the specific type of dental trauma. The output end of the trauma identification module 20 is connected to the input end of the injury grading module 30.

[0027] The injury grading module 30 is used to quantitatively evaluate the identification results output by the trauma identification module 20, and generate a standardized injury grade. The output end of the injury grading module 30 is connected to the input end of the treatment advice generation module 40.

[0028] The treatment advice generation module 40 is used to generate structured treatment advice according to the injury grade output by the injury grading module 30. The output end of the treatment advice generation module 40 is connected to the input end of the human-computer interaction module 50, and is used to show the advice to the user.

[0029] The prognosis risk dynamic assessment module 60 is connected to the human-computer interaction module 50 and the injury grading module 30, and is used to receive initial injury information and subsequent review data. The output end of the prognosis risk dynamic assessment module 60 is also connected to the human-computer interaction module 50, and is used to output a dynamically updated prognosis risk assessment result.

[0030] Reference Figure 2 , Figure 2 is a flowchart of a child dental trauma automatic grading and treatment advice generation method according to an embodiment of the present application. The present application provides a child dental trauma automatic grading and treatment advice generation method, which can comprise the following steps: S1: Acquire image data, clinical parameters and complaint text containing child dental trauma information through the perception acquisition module 10, and perform standardized preprocessing on the data.

[0031] S2: The trauma identification module 20 receives the standardized data, uses a multi-label classification model to identify trauma types such as dental damage, tooth displacement and occlusion relationship disorder, and generates a visual heat map containing damage area positioning information.

[0032] S3: The injury classification module 30 receives the injury recognition result, fuses the injury features, timeliness information and child development information through a hierarchical multi-modal scoring network, calculates the comprehensive injury score and maps it to a standardized injury level.

[0033] S4: The treatment recommendation generation module 40 calls the knowledge graph and rule engine according to the injury level output by S3, generates a preliminary standardized treatment path, and dynamically corrects the path according to the detection result of the concurrent occlusion relationship disorder output by S2, and finally forms a structured treatment recommendation.

[0034] S5: The prognosis risk dynamic assessment module 60 continuously receives the time-series review data uploaded by the user through the human-computer interaction module 50 after the initial diagnosis and treatment, including review images and clinical parameters, and calculates the dynamic prognosis risk assessment result. When the result exceeds the preset threshold, an active warning instruction is generated.

[0035] Reference Figure 1 The perception acquisition module 10 is the data input unit of the automatic classification and treatment recommendation generation system for children's dental trauma, which is bidirectionally connected with the human-computer interaction module 50 to receive the user's operation instructions and data input, and its output end is connected to the injury recognition module 20 for providing standardized multi-modal data.

[0036] The perception acquisition module 10 includes an image acquisition sub-module and a clinical data input sub-module in specific implementation.

[0037] The image acquisition sub-module is used to obtain image or video data of the child's oral cavity through a configured image acquisition interface from devices such as smartphone cameras, intraoral scanners or portable oral cameras. In order to ensure the standardization of input data, an edge computing model running on the terminal device is deployed inside the sub-module to realize interactive shooting guidance. The guidance process includes image quality assessment and acquisition angle assessment.

[0038] Image quality assessment aims to judge the usability of the collected image. Specifically, the system extracts three feature parameters of sharpness, illumination and completeness from the image, and based on these parameters, the image quality score is calculated using the following formula: ; In the formula, is the final calculated image quality score, which is normalized to 0 to 1; is the sharpness feature value, which is obtained by calculating the Laplacian operator variance of the tooth region in the image, and this value reflects the focusing and blurring degree of the image; is the illumination feature value, which is obtained by analyzing the brightness histogram distribution of the tooth region in the image, and this value reflects whether the image is overexposed or underexposed; For completeness feature value, whether the target tooth and its adjacent gingival tissue are completely contained in the image field of view is judged by a preset target detection model; 、 、 are the preset weight coefficients corresponding to the clarity, illumination and completeness respectively.

[0039] The acquisition angle evaluation aims to ensure that all necessary diagnostic perspectives (e.g. labial, lingual / palatal, occlusal) are acquired. This function is implemented by a pre-trained lightweight convolutional neural network (CNN) that is trained to recognize the shooting angle of an image. The network receives as input the image that passed the quality evaluation and outputs a probability vector. The system determines the most likely shooting angle from this probability vector, calculated by the following formula: ; wherein, is the predicted shooting angle category; is the total number of preset standard shooting angle categories; is the probability value corresponding to the th standard shooting angle output by the model.

[0040] In the interactive shooting guidance process, the system first calculates in real time. If this value is lower than the preset quality threshold, the system outputs specific adjustment instructions to the user through the human-computer interaction module 50, such as asking to approach the shooting object or to increase the ambient light. When meets the requirements, the system then calculates and marks the successfully acquired angles on the interface of the human-computer interaction module 50, while prompting the user which remaining angles need to be acquired, until a complete set of multi-perspective images of high quality is obtained.

[0041] The clinical data input submodule is used to collect and structure the basic information and clinical complaints of the patient through the standardized form or voice recognition interface provided by the human-computer interaction module 50. The information includes: patient age, gender, tooth number to be analyzed (following the FDI tooth notation), tooth type (primary or permanent), interval from trauma occurrence to current time, subjective pain visual analog scale (VAS) score, and trauma history, etc.

[0042] After completing image data acquisition and clinical data input, the perception acquisition module 10 uniformly standardizes the above two different modalities of data. This packaging process includes assigning a unique identifier to each case data, adding accurate timestamps to all acquired images, videos and parameters, and integrating all information into a structured data set, which is transmitted to the trauma identification module 20.

[0043] Referring to Figure 1 and Figure 3 , the input of the trauma identification module 20 is connected with the perception acquisition module 10, and the output is connected with the injury grading module 30. The core task is to receive standardized multi-view image data, and concurrently identify one or more types of dental trauma from the data, while extracting feature information for judging key complications (such as occlusion relationship disorder), and combining these identifications and extracted information into a structured identification result.

[0044] In one specific embodiment, a multi-task neural network model based on deep learning is integrated inside the trauma identification module 20. The network architecture of the model takes a convolutional neural network (CNN) as the backbone network, such as a ResNet or EfficientNet structure pre-trained on a large image dataset. The backbone network is responsible for extracting general, hierarchical visual features from the input dental trauma images.

[0045] After the output of the top-layer feature map of the backbone network, the model introduces an attention mechanism module. The attention mechanism (such as channel attention or spatial attention mechanism) performs weighted processing on the feature map extracted by the backbone network, so that the model can allocate more computing resources to feature channels or spatial regions containing key damage information (such as micro-cracks, abnormal tooth color, or displacement profile).

[0046] After the attention mechanism processing, the features are sent to two parallel branch output heads, which perform different tasks respectively: The first branch is a multi-label classification head. The branch is composed of one or more fully connected layers and a final Sigmoid activation layer. It is responsible for identifying a plurality of predefined dental trauma types. For an input case image set, the branch outputs a dental trauma multi-label probability vector. The determination method of the vector is as follows: ; In the formula, is the dental trauma multi-label probability vector; is the total number of predefined dental trauma types, such as enamel fracture, dentin fracture, crown-root fracture, tooth concussion, and tooth subluxation; is the logistic regression output value of the fully connected layer to the th trauma type; is the Sigmoid activation function, which maps each logistic regression output value to the interval (0, 1) to obtain a probability value representing the existence of the th trauma type. When a certain probability value exceeds a preset classification threshold, it is determined that the corresponding trauma type exists.

[0047] The second branch is the occlusion disorder feature extraction head. This branch consists of a fully connected layer, which outputs a single scalar value as the final output. This scalar value is the quantitative score of the occlusion disorder. This score is a quantitative representation of the occlusion status in the high-dimensional feature space. Subsequently, the trauma recognition module 20 generates the occlusion disorder feature vector as claimed in the claims based on this quantitative score, with the following generation rule: Boolean flag: Compare the quantitative score with a pre-set zero-point threshold. If the score is greater than the threshold, the Boolean flag is True, otherwise it is False.

[0048] Quantitative score: Directly use the original scalar value output by this branch. The final generated occlusion disorder feature vector is output to the injury classification module 30 and the treatment recommendation generation module 40 as an independent feature, for more detailed occlusion impact assessment and treatment decision modification.

[0049] To improve the interpretability of the model's decision, the trauma recognition module 20 also includes a damage region visualization unit. When the multi-label classification head identifies a specific trauma type, this unit uses techniques such as Grad-CAM to generate a corresponding damage region heat map. The generation principle of the heat map is as follows: First, calculate the gradient of the model's prediction score for this specific trauma type with respect to the feature map output by the last convolutional layer of the backbone network. Then, use these gradients as weights to perform weighted summation on the feature map to obtain a rough positioning map. This process can be represented by the following formula: ; where ; In the formula, is the original heat map generated for class (i.e., a certain type of trauma); is the prediction score of the model for class ; is the th feature map output by the last convolutional layer; is the value of feature map at spatial position ; is the importance weight of the th feature map for class ; is the total number of pixels (i.e., width x height) of feature map ; is the rectified linear unit activation function, used to filter out negative contributions.

[0050] The generated heat map After up-sampling to the original image size, and superimposed on the original input image in a semi-transparent manner. Referring to Figure 3 Finally, the structured recognition results output by the trauma identification module 20 include the tooth trauma multi-label probability vector, the occlusion disturbance feature vector, and the visualized damage area heat map associated with the identified trauma type, which are transmitted to the downstream module and the human-computer interaction module 50 for display.

[0051] Referring to Figure 1 and Figure 3 The input end of the injury grading module 30 is connected with the trauma identification module 20, and the output end is connected with the treatment suggestion generation module 40. Its function is to convert the discrete multi-label recognition results output by the trauma identification module 20 and the related clinical information into a quantitative and standardized comprehensive injury grade.

[0052] In order to realize this conversion, the injury grading module 30 in the embodiment of the application adopts a hierarchical multi-modal scoring network. This network is not a single deep learning model, but a structured scoring framework for integrating image recognition features, clinical text information, timeliness information and child development stage information for comprehensive quantitative evaluation. The network first calculates the sub-item scores of each dimension independently, and then calculates the comprehensive injury score by weighted summation.

[0053] Specifically, the module first receives the tooth trauma multi-label probability vector and the occlusion disturbance feature vector from the trauma identification module 20, and obtains the timeliness information (trauma occurrence interval) and the child development information (age, tooth type) from the perception acquisition module 10. Based on these multi-modal inputs, the system calculates a comprehensive injury score. The score is determined by the following formula: ; In the formula, is the comprehensive injury score; is the tooth structure damage score; is the tooth displacement damage score; is the occlusion impact score; is the timeliness and development risk score; are the preset weight coefficients of , , , respectively, all of which are positive real numbers and their sum is 1. These weight coefficients are obtained by pre-calibration according to a large number of historical case data and clinical expert guidelines, and reflect the contribution of different damage dimensions to the overall condition.

[0054] The quantitative method of each sub-item score is as follows: Tooth structure damage score ( The score is calculated based on the identification results of tooth structural damage output by the trauma identification module 20. The system maintains a predefined structural damage severity mapping table, which maps each type of tooth structural damage (such as enamel fracture, dentin fracture, and crown-root joint fracture) to a basic severity score. The value is the sum of the baseline severity scores corresponding to all identified structural damage types in the current case.

[0055] Tooth displacement damage score ( This score is calculated based on the identification results regarding changes in tooth position. Similarly, the system also maintains a displacement damage severity mapping table, mapping each type of tooth displacement (such as tooth concussion, tooth subluxation, intrusive or extrusive displacement) to a basic severity score. The value is the sum of the baseline severity scores corresponding to all identified displacement injury types in the current case.

[0056] Occlusal Influence Score ( This score quantifies the impact of dental trauma on a child's occlusal function. Its calculation does not directly rely on classification labels, but rather utilizes the occlusal disorder feature vector output by the trauma recognition module 20. The vector is fed into a pre-trained lightweight regression model (e.g., a small multilayer perceptron), which outputs a single continuous value as... This regression model is trained by learning from a large amount of case data containing information on occlusal status, and its output value can accurately reflect the severity of occlusal interference.

[0057] Timeliness and Developmental Risk Score ( This score is a composite score, determined by both the risk of delayed medical attention and the child's developmental risk. The risk of delayed medical attention is calculated based on the time interval between the occurrence of the injury and the current time, obtained from the perception and acquisition module 10; the longer the time interval, the higher the risk score. For the assessment of the child's developmental risk, the module is configured to intelligently invoke a set of preset, differentiated risk assessment rules based on the child's age and the type of tooth involved (deciduous or permanent). These rules can quantify the specific risks at different physiological stages. For example, for young permanent teeth whose apical foramina have not yet closed, the risk weight of subsequent developmental impact is significantly higher than that of permanent teeth whose roots have fully developed. Conversely, for deciduous teeth nearing replacement, the risk of damage affecting the permanent tooth germ is taken into consideration. It is the result of combining these two risk scores according to preset rules.

[0058] The final comprehensive injury score was calculated. Afterwards, the injury grading module 30 compares the continuous score value with a set of pre-defined thresholds, mapping the continuous score value to a discrete, standardized injury grade (e.g. Grade I - minor, Grade II - moderate, Grade III - severe). The injury grade serves as the final output of the module, and is transmitted to the treatment recommendation generation module 40.

[0059] Referring to Figure 1 and Figure 4 The input of the treatment recommendation generation module 40 is connected to the injury grading module 30 and the trauma identification module 20, and the output is connected to the human-computer interaction module 50. The core decision function of the module is to generate scientific and individualized treatment recommendations based on the quantitative injury grade output by the injury grading module 30 and the complication detection results output by the trauma identification module 20.

[0060] In a specific embodiment of the present application, the module achieves its function through the collaborative work of a knowledge graph and a rule engine.

[0061] The knowledge graph is used to store professional clinical knowledge in the field of dental trauma in a structured manner. The construction method is as follows: Nodes: Define the core entities in the field. Node types include: dental trauma types (such as enamel fracture), injury grades (such as Grade II - moderate), treatment measures (such as tooth reduction, root canal treatment), examination items (such as X-ray examination), drugs (such as antibiotics), complications (such as pulp necrosis), and treatment principles (such as emergency treatment, elective treatment).

[0062] Relationship edges: Define the logical relationship between nodes. Relationship types include: is_level_of, connecting dental trauma types and injury grades; has_standard_procedure, connecting injury grades and treatment measures; requires_test, connecting treatment measures and examination items; may_cause, connecting dental trauma types and complications.

[0063] The rule engine is the decision core of the module, and its working mechanism is to receive data from the upstream module as initial facts (Facts), and execute a series of pre-defined logical rules (Rules) to query the knowledge graph, and finally infer specific treatment recommendations. The inference process includes two steps of standard path inference and dynamic correction.

[0064] In the standard path reasoning step, the rule engine first takes the comprehensive injury grade output by the injury grading module 30 as the key input fact. For example, the rule engine will determine whether the input injury grade is severe grade III. If it is determined to be yes, the engine finds the processing principle node connected to the severe grade III node through the standard processing path relationship in the knowledge graph, and obtains the initial processing measure sequence corresponding to the principle, thereby generating a standardized path set as emergency treatment, which contains a series of initial steps such as (immediate debridement and hemostasis, imaging examination, tooth reduction or fixation).

[0065] After obtaining the standard processing path, the dynamic correction step is entered. After obtaining the standard processing path, the rule engine enters the dynamic correction step. This step introduces the occlusion disorder feature vector output by the trauma recognition module 20, especially the quantitative score representing the severity, as the core basis for executing dynamic correction. This correction logic can adjust the standard path at different granularities according to the score: For example, if the score is only slightly higher than the baseline threshold, indicating a slight occlusion disturbance, the rule engine may only fine-tune the parameters of the tooth fixation node in the path, such as adjusting the recommended fixation time from 2 weeks to 3 weeks.

[0066] However, if the score is significantly higher than the threshold, indicating a severe occlusion disorder, the rule engine will perform more complex correction: after the tooth fixation step, a brand new fine occlusion adjustment processing node is dynamically inserted, and a regular occlusion relationship special review item is forcibly added to the final review plan.

[0067] By executing the above two reasoning steps, the treatment suggestion generation module 40 can consider the macro severity of the condition and the micro specific complications to generate a structured, step-by-step, complete treatment suggestion containing preliminary treatment, drug suggestion, and follow-up review plan. The suggestion is then transmitted to the human-computer interaction module 50 for final presentation.

[0068] Reference Figure 1 The input end of the prognosis risk dynamic assessment module 60 is connected with the injury grading module 30 and the human-computer interaction module 50, and the output end is also connected with the human-computer interaction module 50. The core function of this module is to realize the continuous tracking and assessment of the prognosis risk of children's dental trauma, and to change the traditional one-time diagnosis and treatment assessment mode into a dynamic and closed-loop risk management process covering the entire recovery period.

[0069] The module is activated after the initial diagnosis and treatment is completed, and its workflow is as follows: first, the module receives the initial comprehensive injury score from the injury grading module 30 and take it as the initial value of the prognosis risk. Then, during the whole recovery period, the module receives the user-uploaded follow-up data at pre-set follow-up time points (e.g. 1 week, 1 month, 3 months after surgery) through the human-computer interaction module 50. The follow-up data includes two categories: follow-up image data and structured clinical parameters.

[0070] The prognosis risk dynamic assessment module 60, at each follow-up time point processes the newly received data and combines the historical risk status to calculate a current dynamic prognosis risk assessment result. The calculation of the assessment result is determined by the following formula: ; wherein, is the current dynamic prognosis risk assessment result, the initial state of the value is the comprehensive injury score calculated by the injury grading module 30 at the initial diagnosis and treatment is initialized; is the prognosis risk assessment result at the previous time point; is the image dimension risk value, which is calculated by a special image analysis model trained to detect subtle imaging signs of poor prognosis, such as the appearance or enlargement of periapical shadow, internal or external resorption of the root, etc. is the clinical parameter risk value, the system converts the user feedback parameters (such as whether the tooth color is dark, whether there is tenderness, whether there is looseness) into corresponding scores through a pre-set clinical sign risk score mapping table, and then sums the weighted values to obtain ; is the historical risk decay coefficient, which is a pre-set coefficient with a value range of 0 to 1, used to adjust the influence degree of the historical risk status on the current assessment result. A higher value indicates that the historical risk continues to have a greater impact, and vice versa, indicating that the current assessment is more focused on the latest clinical performance; is the weight coefficient of the image dimension risk value; is the weight coefficient of the clinical parameter risk value; is the basic risk constant.

[0071] After calculating the current dynamic prognosis risk assessment result , the module performs active early warning judgment. This mechanism is achieved by comparing the value of with one or more pre-set risk warning thresholds (e.g. ). When it is determined that the value of exceeds the warning threshold When the prognosis risk dynamic assessment module 60 detects a significant increase in the risk of pulp necrosis, it generates an active warning instruction. This instruction is sent to the human-computer interaction module 50, which presents a clear risk prompt to the user (doctor or parent), such as detecting an increased risk of pulp necrosis, suggesting immediate appointment for X-ray examination, so as to achieve early detection and intervention of potential complications.

[0072] With reference to Figure 1 and Figure 5 , the human-computer interaction module 50 is the only interface for the entire system to exchange information with the end user. Its input is connected to the processing suggestion generation module 40 and the prognosis risk dynamic assessment module 60 to receive the results of system analysis and decision-making; its output is connected to the perception acquisition module 10 and the prognosis risk dynamic assessment module 60 to return the user input data to the corresponding processing module.

[0073] A core feature of the human-computer interaction module 50 in the embodiment of the present application is its ability to adaptively present information to users with different roles. The module contains an information adaptation logic unit inside. When the user first uses or logs in, the system will determine his / her role as a doctor or a parent / guardian. Then, when receiving the same structured processing suggestion from the processing suggestion generation module 40, the logic unit will execute different content filtering, term conversion and format reconstruction rules according to different user roles, and output two interfaces with completely different levels of detail and professionalism.

[0074] For users with the role of a doctor, the human-computer interaction module 50 presents a professional diagnosis and treatment interface. The design of this interface is based on the principles of comprehensive information and accurate data. It will fully display all the information output by the upstream modules, including: The tooth trauma multi-label probability vector, accurate medical diagnosis terminology (such as non-complex fracture of the dental crown with embedded tooth dislocation) and damage area visualization heat map for auxiliary diagnosis output by the trauma identification module 20.

[0075] The quantitative scores output by the injury grading module 30, including the comprehensive injury score and its specific numerical value.

[0076] The complete diagnosis and treatment plan output by the processing suggestion generation module 40, which contains detailed steps without simplification, clearly lists specific treatment measures, recommended examination items (such as X-ray type), drug names, doses and accurate review schedule.

[0077] The numerical value and historical change curve of the dynamic prognosis risk assessment result output by the prognosis risk dynamic assessment module 60.

[0078] For the user in the role of a parent / guardian, the human-machine interaction module 50 presents a home care guidance interface. The design of this interface is based on the principles of simplicity and operability. The information adaptation logic unit converts the professional diagnosis and treatment plan into a pictorial care guide, which includes: Describes the injury using non-specialized everyday language (e.g., converts an enamel dentin fracture into a tooth with a small piece knocked off, revealing the yellow part inside).

[0079] Replaces the quantitative risk score with a qualitative state description (e.g., presents a high risk score as a situation that is relatively urgent, and advises seeking medical attention as soon as possible).

[0080] Simplifies the complex treatment plan into several key items that the parent needs to perform or cooperate with, such as dietary guidance: give the child soft or soft food for the next week, oral hygiene: guide the child to gently clean the injured tooth with a soft toothbrush, etc., and provide explanatory pictures or animations for each item.

[0081] Converts the professional review plan into a simple calendar reminder and automatically sets a reminder.

[0082] In addition, the human-machine interaction module 50 also undertakes the function of data transmission and implementation of closed-loop follow-up. Under the driving of the prognosis risk dynamic assessment module 60, the module will actively push notifications to the parent end user at the preset review time point. The user can perform the following operations through the interface provided by the module: Review image upload: the interface will call the camera of the terminal device and activate the interactive shooting guidance function of the perception acquisition module 10 to assist the parent in shooting multiple-angle review images that meet the quality requirements.

[0083] Structured questionnaire filling: the interface presents a guided electronic questionnaire containing a series of questions described in simple language (e.g., has the child's tooth color and surrounding healthy teeth changed? Does the child cry when eating?), and the user's choices are automatically converted by the module into standardized parameters for calculating the clinical dimension risk value .

[0084] When the user completes data submission, the human-machine interaction module 50 is responsible for packaging and securely transmitting these newly acquired images and parameters, along with the user identifier and timestamp information, to the prognosis risk dynamic assessment module 60 for updating the risk state. When the module generates a new warning instruction, the human-machine interaction module 50 receives the instruction and updates the user interface in real time, thereby completing a closed-loop follow-up process of data acquisition-risk assessment-warning feedback.

[0085] The child tooth trauma automatic grading and treatment suggestion generation system provided by the application adopts a deployment architecture of edge computing and cloud server cooperative processing, and integrates strict data security and privacy protection mechanisms in the design.

[0086] In order to realize low-delay user interaction and basic functions in the scene of unstable or unavailable network connection, part of the intelligent computing tasks of the system are deployed on the user terminal device (such as a smart phone) for execution. Specifically, the image quality evaluation model and the collection angle evaluation model inside the perception collection module 10 for realizing interactive shooting guidance are edge computing models that have been subjected to lightweight processing (for example, through model quantization or pruning technology). This enables the user to obtain immediate feedback and guidance without network communication during the process of collecting image data, ensuring the efficiency and quality of the data collection link.

[0087] The system has offline processing capability. The user can start the application program in a completely offline state and complete the entire workflow of the perception collection module 10, including shooting a set of multi-view images that meet the quality and angle requirements, and filling in all the required clinical parameter questionnaires. During this process, all collected data are temporarily and securely stored as a structured data package on the user terminal device. When the terminal device regains network connection, the system will automatically upload the data package to the cloud server through an encrypted channel to start subsequent computationally intensive processing procedures such as trauma identification and injury grading. This storage and forwarding mechanism ensures the continuity of data collection and is not limited by the network environment.

[0088] The system implements multi-level security and privacy protection measures throughout the entire life cycle of data processing.

[0089] Data encryption: all data transmitted between the user terminal and the cloud server are encrypted through the Transport Layer Security (TLS) protocol to prevent eavesdropping or tampering during transmission. All user data stored in the cloud server database, especially data containing personal identity information (PII) and medical health information (PHI), are statically encrypted using algorithms such as Advanced Encryption Standard (AES-256) for storage.

[0090] Data anonymization processing: before data is used for algorithm model training or statistical analysis, the system will perform strict anonymization processing. This processing completely separates information that can identify individual identity (such as name, contact information) from clinical medical data, and uses a unique, irreversible anonymized identifier to associate the user's medical data. This ensures that specific individuals cannot be traced during data analysis and research, thereby protecting user privacy.

[0091] Permission management: The system establishes a role-based access control (RBAC) mechanism. The user roles in the system are clearly divided into doctors, parents / guardians, and system administrators, etc. Different roles have strictly defined data access and operation permissions. For example, the parent / guardian role can only access and manage the medical data of its associated minor children; the doctor role can only access the patient data within its scope of diagnosis and treatment relationship; and the system administrator role has system management permissions, but the access to specific medical data is strictly limited and audited.

[0092] Log audit: The system records detailed logs for all key operations, especially for access, modification, and deletion of sensitive data. The audit log contains information such as operator identity, operation time, operation type, and data object being operated. This mechanism provides reliable evidence for post-tracing, security event analysis, and compliance review.

[0093] The complete flow of the child dental trauma intelligent grading and dynamic processing decision assistance method provided by the present application will be described below in conjunction with the accompanying Figure 2 to Figure 5 A hypothetical case will be used to illustrate the complete flow of the child dental trauma intelligent grading and dynamic processing decision assistance method provided by the present application. This flow corresponds to the S1 to S6 steps shown in Figure 2 and demonstrates the collaborative working mode between the system modules.

[0094] Case background: An 8-year-old child suffered a maxillary central incisor (tooth number 11) trauma due to a fall. Two hours after the injury, the child's parent used the system of the present application to perform a preliminary assessment.

[0095] Referring to Figure 5 , the parent starts the application on his smartphone, and the human-computer interaction module 50 presents a guide interface. The parent selects the I am a parent role and starts a new case. According to the interface prompts, the parent inputs the child's information: age 8 years old, gender male, and tooth type permanent. Subsequently, under the interactive shooting guidance of the perception acquisition module 10, the parent takes pictures of the child's tooth number 11 from the labial side, lingual side, and incisal end. In this process, the edge computing model evaluates the image clarity, lighting, and integrity in real time and prompts the parent to adjust the distance and angle until the image meeting the quality requirements is collected. Finally, the parent answers questions such as the trauma occurring time being 2 hours ago and the child feeling the tooth being somewhat elongated and not daring to touch, etc. in a structured questionnaire. All the collected data are packaged into a standardized data package and uploaded through an encrypted channel.

[0096] After receiving the data package, the trauma recognition module 20 in the cloud starts to analyze the multi-view images using the multi-task neural network model inside it. Referring to Figure 3, the backbone network of the model extracts visual features, which are focused by the attention mechanism, and then two branches output heads work in parallel. The probability vector output by the multi-label classification head In this case, the probability values of both the tooth non-complex crown fracture and the tooth extrusive subluxation categories exceed the preset threshold. Meanwhile, the occlusion disorder feature extraction head generates a feature vector , the value of which reflects the occlusion interference that may be caused by tooth elongation. In addition, the damage area visualization unit generates a heat map corresponding to the recognition result, highlighting the broken line of the tooth and the displacement area of the gum edge on the image.

[0097] The injury classification module 30 receives the recognition labels (crown fracture, extrusive subluxation) and feature vectors output by S2, as well as the clinical information (8 years old, permanent teeth, 2-hour delay) collected by S1. The hierarchical multi-modal scoring network within the module begins to calculate the scores of each item: According to the severity mapping table of crown fracture, an intermediate score is obtained. According to the severity mapping table of extrusive subluxation, a higher score is obtained. Through the regression model analysis of the feature vector , a score indicating the presence of moderate occlusion interference is output. Taking into account the 2-hour delay in treatment and the high risk of development of young permanent teeth (aperture not closed) at the age of 8, a higher score is obtained.

[0098] Subsequently, the module calculates the comprehensive injury score according to the formula. Due to the high scores of multiple items, the final comprehensive score falls into the severe Ⅲ interval.

[0099] Referring to Figure 4 , the treatment suggestion generation module 40 takes the Ⅲ severe as input facts, and starts the rule engine. The standard path reasoning first matches to this level, and extracts the standardized treatment path under the principle of emergency treatment from the knowledge graph, including steps such as X-ray examination, debridement and disinfection, tooth reduction and elastic fixation. Subsequently, the dynamic revision logic is started, and it is found that the occlusion impact score exceeds the threshold, so the step of checking and adjusting the early contact point is automatically inserted after tooth reduction, and the monitoring of occlusion relationship is emphasized in the review plan. Finally, a structured treatment suggestion containing emergency treatment, reduction and fixation, occlusion adjustment and follow-up review plan is completely generated.

[0100] The prognosis risk dynamic assessment module 60 is activated, taking the high score calculated by S3 as the initial value Meanwhile, based on the follow-up plan in the processing suggestions generated by S4, the module created a follow-up task queue for the child in the system and set a plan to push follow-up reminders to the parents at key time points such as 1 week, 1 month, and 3 months.

[0101] The human-computer interaction module 50 receives the final processing suggestions and risk status. Since the user role is a parent, the information adaptation logic unit is activated, converting the results into a parent-side care guide interface (see...). Figure 5 The interface prominently displays a red warning at the top: "This is an emergency. Please take your child to the vet immediately." Below, it lists pre-visit precautions with illustrations (such as holding the teeth and avoiding biting), and provides one-click navigation to nearby dental emergency rooms. When a professional doctor logs into the system using their account to access the case, the human-computer interaction module presents a completely different professional interface, detailing all identification tags, quantitative scores, heatmaps, complete treatment steps, and the set follow-up plan. This concludes a complete decision support process.

[0102] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart grading and dynamic processing decision assistance system for children's dental trauma, characterized by, The system comprises: a perception acquisition module for acquiring multi-modal data including images of affected teeth; a trauma identification module for analyzing images of affected teeth in the multi-modal data to identify the type of dental trauma and detect the presence of concurrent occlusal relationship disorders, and combining the type of dental trauma and the detection results of concurrent occlusal relationship disorders into a structured identification result; a trauma grading module for receiving the structured identification result and combining other information in the multi-modal data to comprehensively score the trauma through a hierarchical multi-modal scoring network and output a standardized trauma grade; a treatment recommendation generation module for generating a structured treatment recommendation based on the standardized trauma grade, and particularly according to the detection results of concurrent occlusal relationship disorders in the structured identification result; a human-computer interaction module connected with the perception acquisition module and the treatment recommendation generation module respectively for information input and treatment result display of the user terminal, specifically for: receiving the multi-modal data input by the user for the perception acquisition module to acquire, and displaying the structured treatment recommendation output by the treatment recommendation generation module to the user terminal.

2. The intelligent grading and dynamic processing decision-aiding system for dental trauma of children according to claim 1, characterized in that, The perception acquisition module comprises: deploying and running an edge computing model locally on the mobile terminal, which analyzes the camera view screen of the mobile terminal in real time to evaluate the image acquisition quality and angle of the current screen; outputting real-time interactive guidance instructions to the user terminal based on the evaluation results, which are used to guide the user to adjust the shooting behavior until the edge computing model determines that the images of affected teeth meet the preset standardized acquisition requirements.

3. The intelligent grading and dynamic processing decision-aiding system for dental trauma of children according to claim 1, characterized in that, The trauma identification module comprises: a multi-label classification model having a shared feature extraction backbone network and at least two parallel output branches for classification and feature extraction respectively, and using the backbone network and output branches to concurrently analyze and output a structured identification result from the images of affected teeth, and generate a damage area heat map for representing the corresponding position of the type of dental trauma in the images of affected teeth; generated by the classification output branch of the multi-label classification model, used to indicate the probability of the existence of each of a plurality of preset types of dental trauma; generated by the feature extraction output branch of the multi-label classification model, used to represent the occlusal state of the affected teeth in the form of numerical feature data, wherein the values of the feature data are associated with the clinical severity of occlusal disorders.

4. The intelligent grading and dynamic processing decision-aiding system for dental trauma of children according to claim 3, characterized in that, The trauma grading module comprises: using the structured identification result output by the trauma identification module as the core input, and combining the time-sensitive and developmental information in the multi-modal data to calculate a comprehensive trauma score; wherein the trauma grading module further comprises: According to the tooth type and the age of the child contained in the development information, preset differentiated risk assessment rules for different tooth types and development stages are called to calculate the timeliness and development risk score; and the comprehensive injury score The calculation formula of the comprehensive injury score includes: ； In the formula, is a comprehensive injury score; is a dental structure injury score; is a dental displacement injury score; is a bite influence score; is a timeliness and development risk score; respectively correspond to preset weight coefficients of , , , .

5. The intelligent grading and dynamic processing decision-aiding system for dental trauma of children according to claim 1, characterized in that, The treatment recommendation generation module comprises a knowledge graph and a rule engine, wherein, the knowledge graph is used to structure standard dental trauma treatment guidelines and clinical expert experience into a graph data structure comprising medical concept nodes and semantic relationship edges; the rule engine runs based on the knowledge graph and generates treatment recommendations through the following two steps: Generating a standard path: the rule engine infers a standard treatment path matching a standardized injury level by traversing the medical concept nodes and semantic relationship edges in the graph data structure; Performing dynamic revision: when the detection result about the concurrent occlusal relationship disorder in the structured recognition result is true, the rule engine further includes revision logic for inserting additional complication treatment nodes in the standard treatment path or adjusting the treatment parameters of the existing nodes in the standard treatment path according to the quantitative scores in the occlusal disorder feature vector, to generate a final structured treatment suggestion.

6. The intelligent grading and dynamic processing decision-aiding system for dental trauma of children according to claim 1, characterized in that, The human-computer interaction module includes: Building a professional interface for doctors and a guidance interface for parents; And the structured treatment suggestion output by the treatment suggestion generation module is respectively adapted to generate a detailed diagnosis and treatment plan for professionals and a graphic nursing guide for non-professionals, and then the detailed diagnosis and treatment plan is displayed on the professional interface for doctors, and the graphic nursing guide is displayed on the guidance interface for parents.

7. The intelligent grading and dynamic processing decision-aiding system for dental trauma of children of claim 1, wherein, The system further includes: A prognosis risk dynamic evaluation module connected with the human-computer interaction module, for receiving time-series review data uploaded by the user during follow-up, the time-series review data including current tooth image and clinical review parameters; The prognosis risk dynamic evaluation module is further used to perform feature-level difference comparison between the current tooth image and the historical tooth image to quantify the risk indicators in the image dimension, and fuse the quantified risk indicators in the image dimension, the clinical review parameters, and the prognosis risk evaluation result at the last time point, to update and output the current dynamic prognosis risk evaluation result.

8. The intelligent grading and dynamic processing decision-aiding system for dental trauma of children according to claim 7, characterized in that, the current dynamic prognosis risk assessment result the calculation formula Including: ； wherein is the current dynamic prognosis risk assessment result; is the previous time point prognosis risk assessment result; is the image dimension risk value; is the clinical parameter risk value; is the historical risk decay coefficient; is the weight coefficient of the image dimension risk value; is the weight coefficient of the clinical parameter risk value; is the base risk constant; wherein the image dimension risk value is obtained by feature-level differential comparison between the current dental image and the dental image at the last time point, the comparison including at least one of the following: crown color change, gingiva morphology and color change, and periapical radiographic change.

9. The intelligent grading and dynamic processing decision-aiding system for dental trauma of children according to claim 8, characterized in that, The prognosis risk dynamic evaluation module is further configured to trigger an active early warning instruction when the current dynamic prognosis risk evaluation result When the preset early warning threshold is exceeded, an active early warning instruction is triggered, and early warning information is pushed to a user terminal through the human-computer interaction module.

10. A method for intelligent grading and dynamic processing decision assistance of dental trauma in children, applied to the system of any one of claims 1-9, characterized in that, The method includes the following steps: S1, receiving multi-modal data including tooth image through a user terminal; S2, analyzing the tooth image in the multi-modal data to identify the type of dental trauma and detect whether there is concurrent occlusal relationship disorder, and combining the type of dental trauma and the detection result of the concurrent occlusal relationship disorder into a structured recognition result; S3, receiving the structured recognition result and combining other information in the multi-modal data, and comprehensively scoring the injury through a hierarchical multi-modal scoring network to output a standardized injury level; S4, generating a structured treatment suggestion based on the standardized injury level and according to the detection result of the concurrent occlusal relationship disorder in the structured recognition result; S5, displaying the structured treatment suggestion on the user terminal.

Images