A method and system for predicting and recording dental trauma

Abstract

This application discloses a dental trauma prediction and information recording system. This application relates to the field of oral medicine technology and aims to solve the technical problems of inaccurate extraction of tooth force signals caused by noise interference from saliva flow, chewing movements, etc. in the oral environment, and the inability to predict, warn, and retain information about dental trauma in real time, which makes it difficult to support clinical diagnosis and treatment. Its technical solution includes a wearable sensing module, an algorithm operation module, an early warning module, a storage module, a communication module, and a terminal display module. The wearable sensing module is a brace-like device that integrates multiple sensors to collect multimodal mechanical signals. The algorithm operation module uses a microcontroller as the core to complete the entire process of algorithm operation. All modules work together to realize signal acquisition, processing, prediction, early warning, storage, and visualization display.

Description

Technical Field

[0001] This application relates to the field of oral medicine technology, specifically to a method and system for predicting and recording dental trauma. Background Technology

[0002] In current clinical dental practice, dental trauma (such as falls, sports injuries, traffic accidents, etc.) is a common emergency in children and adolescents, and the quality of its diagnosis and treatment highly depends on the accurate acquisition of key mechanical parameters such as the direction, force, and time of the trauma at the moment of occurrence. Current mainstream methods fall into two categories: one is based on the doctor's experience in judging the injury according to the patient's complaints and imaging examinations (X-rays, CBCT); the other is that some studies use fixed intraoral sensors or mandibular patches to collect force signals, record impact events through threshold triggering, and analyze damage characteristics through post-incident playback. In recent years, with the development of MEMS sensing and edge AI technologies, some devices have attempted to integrate accelerometers and pressure sensors to achieve preliminary data acquisition; some algorithm platforms have also introduced support vector machines or LSTM models to classify and identify static occlusal force data. These technologies have already formed a certain application foundation in dental laboratories and digital orthodontic assessments, constituting the technical prototype of auxiliary diagnosis for dental trauma.

[0003] However, the aforementioned existing technologies still have significant limitations: On the one hand, the patient's chief complaint information heavily relies on their memory and expressive abilities, resulting in a distortion rate as high as 68% in sudden, high-stress scenarios (such as children, comatose patients, and multilingual individuals), leading to misjudgment of direction and force, directly affecting the determination of injury type (such as crown fracture vs. root fracture); on the other hand, existing wearable acquisition devices have not designed anti-interference mechanisms for the dynamic oral environment, with non-traumatic signals such as saliva flow, chewing artifacts, and respiratory vibrations accounting for over 40%, causing the failure to extract effective mechanical features; in addition, multimodal signals have long been in a state of parallel acquisition—independent storage—manual comparison, lacking a unified temporal alignment and physical meaning-driven feature fusion logic, making it impossible to support real-time prediction and second-level early warning; more importantly, none of the existing algorithms have embedded the grading criteria in the IADT International Dental Trauma Guidelines, nor have they been linked with clinical treatment pathways, making it difficult to output quantitative results that can guide treatment. Therefore, there is an urgent need for a method for predicting and recording information on the entire process of dental trauma that is oriented towards the real dynamic oral environment, has adaptive anti-interference capabilities, integrates three-level targets of direction, force, and damage type, and conforms to clinical guidelines. Summary of the Invention

[0004] This application provides a method and system for predicting and recording dental trauma. It solves the technical problems of inaccurate extraction of tooth force signals caused by noise interference from saliva flow and chewing movements in the oral environment, and the inability to predict, warn, and retain information about dental trauma in real time, which makes it difficult to support clinical diagnosis and treatment.

[0005] To achieve the above objectives, this application provides the following technical solution: This application proposes a method for predicting and recording information on dental trauma, including: In a first aspect, this application provides a dental trauma prediction and information recording system, including a wearable sensing module and an algorithm operation module connected in sequence. The algorithm operation module communicates bidirectionally with an early warning module, a storage module, and a communication module, respectively. The communication module communicates bidirectionally with a terminal display module. The wearable sensing module has a brace-like structure and integrates a three-axis accelerometer, a three-axis gyroscope, a miniature piezoresistive pressure sensor, and a capacitive contact trigger sensor to collect multimodal mechanical signals of force on teeth in the oral cavity. The algorithm operation module is based on a microcontroller with a dental trauma prediction and information recording algorithm, and completes the entire process of data preprocessing, direction and force prediction, and damage parameter identification. The early warning module includes a built-in miniature vibrator in the braces and a portable audible and visual alarm, providing dual early warning for high-risk injuries.

[0006] As a further improvement to this application, it also includes: The storage module adopts a dual architecture of local and cloud distributed storage for encrypted storage of dental trauma-related data; The communication module adopts dual-mode communication to realize real-time data transmission between modules; The terminal display module is used to visually display the algorithm output parameters and support data query and export.

[0007] As a further improvement to this application, the data encrypted and stored by the storage module includes early warning records, trauma records, and raw sensor data; wherein, the early warning records include time, force direction, and predicted force, and the trauma records include time, predicted force, degree of trauma, and type of injury.

[0008] As a further improvement to this application, the wearable sensing module has a triaxial accelerometer; specifically, it is an MPU-6050 with a local storage capacity of 128MB. In the early warning module, the vibration frequency of the micro vibrator is 100-300Hz, and the sound pressure level of the portable audible and visual alarm is ≥80dB.

[0009] Secondly, this application provides a method for predicting and recording dental trauma, applied to the aforementioned system for predicting and recording dental trauma, comprising the following steps: S1. Collect raw signals of triaxial acceleration, triaxial angular velocity, and multimodal mechanical stress on teeth in the oral cavity through a wearable sensing module; S2. Adaptive Kalman filtering and sliding window smoothing algorithms are used to process the original signal and extract the effective mechanical signal; S3. Based on the preprocessed effective signal, extract multi-dimensional features and optimize them. Simultaneously predict the force direction and intensity of dental trauma using an improved support vector machine algorithm. S4. If the predicted intensity exceeds the safety threshold, a fusion algorithm of decision tree and Bayesian classifier is used, combined with adaptive threshold correction rules, to identify the degree of trauma and injury type. S5. If the injury is determined to be of high risk, an early warning will be activated through the early warning module.

[0010] As a further improvement to this application, the implementation process of the adaptive Kalman filter described in S2 includes: Define the state vector X=[a x ,a ,a z ,ω x ,ω ,ω z [,p,φ], where a x / a / a z For triaxial acceleration, ω x / ω / ω z Here, φ represents the triaxial angular velocity, p represents the pressure amplitude, and φ represents the signal phase; the observation equation is Z = HX + V, where H is the observation matrix and V is the observation noise. By calculating the estimated value of the observation noise covariance matrix R in real time, the Kalman gain K is dynamically adjusted. When the signal changes abruptly, the gain is reduced to preserve peak characteristics, and when the signal is stable, the gain is increased to enhance noise suppression. Based on the state prediction value at the previous moment k | k-1 Compared with the current observation value Z k Calculate the optimal estimate k = k | k-1 +K k (Z k -H k | k-1 And update the error covariance matrix P. k .

[0011] As a further improvement to this application, the specific process of smoothing the sliding window described in S2 is as follows: A sliding window with a length in milliseconds (e.g., 50ms) is used to process the filtered signal. The output value is calculated using a weighted average method within the window. The weight of the data in the center of the window is set to 0.3, and the weight of the data on both sides decreases by 0.05.

[0012] As a further improvement to this application, the process of multi-dimensional feature extraction is as follows: Extracting peak value A from acceleration signal p Peak time T p Rising slope K a Root mean square value R ams Four time-domain features; extraction of peak angular acceleration α from angular velocity signal p Two features are identified: the integral value of angular displacement θ; and the peak pressure P is extracted from the pressure signal. p and pressure distribution entropy E p The two features are combined to construct an 8-dimensional feature vector F=[A p ,T p ,K a ,R ams ,α p ,θ,P p E p ]; The process of extracting and optimizing multi-dimensional features based on the preprocessed effective signal is as follows: First, min-max normalization is used to map the 8-dimensional feature vector to the [0,1] interval to eliminate the difference in dimensions; then, a weighted coefficient matrix W=diag[0.3,0.1,0.1,0.1,0.1,0.1,0.3,0.1] is introduced to weight the feature vector, thereby strengthening the influence of the core force characteristics of the peak acceleration and peak pressure.

[0013] As a further improvement to this application, the construction process of the improved support vector machine algorithm is as follows: The kernel function K(x,y)=exp(-||xy||² / (2σ²)) is adopted, and the kernel parameter σ and the penalty factor C are optimized by grid search method, where the search range of σ is [0.01,10] and the search range of C is [0.1,100]. The direction classification sub-model calculates the cosine similarity between the feature vector and six direction templates (horizontal, vertical, 45°, 135°, 225°, and 315°), and outputs the direction with the highest similarity as the prediction result. The force regression sub-model maps the feature vector to a specific force value F through SVM regression. p This allows for the quantification of intensity.

[0014] As a further improvement to this application, the primary classification process of the decision tree is as follows: using the strength value F p , rate of change of force dF p / dt, force direction D, and pressure distribution range S are the decision nodes, and the force threshold F is set. p ≤5N is considered a minor injury. <F p ≤20N is considered moderate trauma, F p >20N indicates severe trauma, completing the preliminary classification of trauma severity; simultaneously, based on the direction and intensity of force, a preliminary candidate set of injury types is determined; The optimization process of the Bayesian classifier is as follows: the output of the decision tree is used as the prior probability P(Y|X1), and combined with the preset damage type-mechanical feature mapping library, the posterior probability P(Y|X1,X2)∝P(X2|Y)P(Y|X1) is calculated, where X2 is the optimized 8-dimensional feature vector and Y is the damage type; the damage type with the highest posterior probability is output as the final result; the damage type-mechanical feature mapping library is constructed through massive clinical cases and simulation experiments, and the damage types include coronal fracture, root fracture, subluxation, and complete dislocation; The specific rule for the adaptive threshold correction is as follows: based on the user's age and tooth type, a linear correction function F is used. th '=F th ×k dynamically adjusts the threshold of the decision node, F th F is the original threshold. th ' is the corrected threshold, and k is the correction coefficient; where, for children k=0.7-0.9, for adults k=1.0; for deciduous teeth k=0.6-0.8, for permanent teeth k=1.0.

[0015] This application also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for predicting and recording dental trauma.

[0016] This application also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for predicting and recording dental trauma.

[0017] This application also proposes a computer program product, which includes computer instructions that instruct a computer to execute the above-described method for predicting and recording dental trauma.

[0018] Compared with the prior art, this application has the following advantages: To address the shortcomings of existing technologies, such as subjective and inaccurate patient reports of dental trauma and the lack of dynamic mechanical signal analysis algorithms adapted to the oral environment, this application provides a method for predicting and recording dental trauma. Simultaneously, it constructs a system based on this algorithm to achieve accurate prediction of the direction and force of dental trauma and objective recording of key trauma information, providing reliable patient reports for clinical treatment. The wearable dental brace-style sensing module adapts to the oral environment and integrates multiple sensors to comprehensively collect multimodal mechanical signals of tooth stress, providing accurate and comprehensive raw data support for subsequent signal processing and algorithm calculations. The algorithm calculation module, with a microcontroller at its core, can efficiently complete the entire algorithm calculation process, automating signal preprocessing, direction and force prediction, and damage parameter identification, thus improving the timeliness and accuracy of dental trauma prediction. Multiple modules work together, with dual early warnings to promptly remind users to avoid high-risk injuries, a dual storage architecture to securely retain relevant data, dual-mode communication to ensure real-time data transmission, and a visual display on the terminal for easy viewing by users and access to diagnostic and treatment information by doctors. This effectively solves the problems of lagging dental trauma prediction, insufficient information retention, and difficulty in supporting clinical diagnosis and treatment in existing technologies, improving the intelligence level and clinical practicality of oral health protection. Attached Figure Description

[0019] Figure 1 The logical relationship between the modules of the dental trauma prediction and information recording system of this application is shown; Figure 2 This is a flowchart of the method for predicting and recording dental trauma in this application. Detailed Implementation

[0020] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0021] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0022] Dental trauma is a common clinical condition in oral medicine, and the accuracy of its diagnosis and treatment highly depends on the patient's core complaints, such as the direction, force, and timing of the injury. However, current technologies rely heavily on the subjective descriptions of the patient, which has significant limitations: the impact and emotional stress at the time of the injury can lead to deviations in the perception of force parameters; and special populations such as children and those with impaired consciousness cannot effectively express their condition, making it difficult for doctors to quickly and accurately determine the type and extent of the injury.

[0023] The inventors have discovered that current equipment in the field of oral healthcare, whether focusing on imaging detection and diagnosis of post-traumatic injuries or on routine oral health monitoring, cannot predict and warn of injuries before they occur, and lacks the ability to accurately capture and record dynamic parameters of the trauma process. Existing data processing algorithms are mostly applied to static image analysis, which is difficult to adapt to the mechanical signal analysis in the dynamic environment of the oral cavity, and cannot meet the needs of real-time prediction and parameter quantification of dental trauma. This application provides an algorithm and related system that adapts to the complex oral environment, can accurately predict the direction and force of dental trauma, and simultaneously complete the quantitative recording of trauma information.

[0024] The first objective of this application is to provide a system for predicting and recording information about dental trauma, such as... Figure 1 As shown, it includes a wearable sensing module, an algorithm processing module, an early warning module, a storage module, a communication module, and a terminal display module; the algorithm processing module is equipped with a dental trauma prediction and information recording algorithm, which is used to process the mechanical signals collected by the wearable sensing module.

[0025] The wearable sensing module has a brace-like structure and integrates a three-axis accelerometer, a three-axis gyroscope, a pressure sensor, and a contact trigger sensor.

[0026] Specifically, a dental trauma prediction and information recording system includes a wearable sensing module and an algorithm operation module that are connected in sequence. The algorithm operation module communicates bidirectionally with an early warning module, a storage module, and a communication module, respectively. The communication module communicates bidirectionally with a terminal display module. The wearable sensing module has a brace-like structure and integrates a three-axis accelerometer, a three-axis gyroscope, a miniature piezoresistive pressure sensor, and a capacitive contact trigger sensor to collect multimodal mechanical signals of the force exerted on the teeth in the oral cavity. The algorithm computing module is based on the STM32H743 microcontroller and is equipped with dental trauma prediction and information recording algorithms to complete the entire process of data preprocessing, direction and force prediction, and damage parameter identification. The early warning module includes a built-in miniature vibrator in the braces and a portable audible and visual alarm for dual early warning in case of high-risk trauma. The storage module adopts a dual architecture of local Flash + cloud distributed storage for encrypted storage of dental trauma-related data; The communication module adopts Bluetooth 5.2 + 5G dual-mode communication to realize real-time data transmission between modules; The terminal display module includes a mobile APP and a doctor's workstation terminal, which are used to visually display algorithm output parameters and support data query and export.

[0027] The second objective of this application is to provide an algorithm for predicting and recording dental trauma, including an adaptive anti-interference data preprocessing algorithm, a multi-feature fusion algorithm for direction and force prediction, and a decision fusion algorithm for damage parameter identification. The adaptive anti-interference data preprocessing algorithm uses a combination of adaptive Kalman filtering and sliding window smoothing to extract effective mechanical signals from the oral cavity. The direction and force prediction algorithm uses an improved SVM to perform multi-feature fusion analysis, outputting the direction of the trauma and quantified force. The damage parameter identification algorithm uses a fusion of decision tree and Bayesian classifier to identify the degree and type of trauma. The adaptive Kalman filter dynamically adjusts the Kalman gain by estimating the observation noise covariance matrix in real time, and the sliding window uses a weighted average window with a length of 50ms. The direction and force prediction algorithm extracts an 8-dimensional feature vector, introduces a weighted coefficient matrix to enhance the influence of peak pressure and peak acceleration, and uses an improved SVM with an RBF kernel function to achieve direction classification and force regression.

[0028] Specifically, the aforementioned method for predicting and recording dental trauma is applied to a system for predicting and recording dental trauma, such as... Figure 2 As shown, the specific steps include: S1. Signal Acquisition: The wearable sensing module acquires raw signals of triaxial acceleration, triaxial angular velocity, and multimodal mechanical stress on teeth in the oral cavity. S2. Adaptive anti-interference data preprocessing: The original signal is processed by a combination algorithm of adaptive Kalman filtering and sliding window smoothing to extract the effective mechanical signal; S3. Direction and force prediction of multi-feature fusion: Based on the preprocessed effective signal, multi-dimensional features are extracted and optimized, and the force direction and force of dental trauma are predicted simultaneously through an improved support vector machine algorithm. S4. Damage parameter identification through decision fusion: If the predicted force exceeds the safety threshold, a fusion algorithm of decision tree + Bayesian classifier is used, combined with adaptive threshold correction rules, to identify the degree of trauma and damage type. The identification criteria refer to the International Association for Dental Trauma (IADT) guidelines. S5. Warning Trigger: If the injury is determined to be of high risk, a dual warning will be triggered through the warning module.

[0029] Furthermore, preferably, it also includes: S6. Data storage and transmission: Dental trauma-related data is encrypted and stored through the storage module, and the algorithm output parameters are transmitted to the terminal display module through the communication module; S7. Visualization: The algorithm output parameters are visualized through the terminal display module, and data query and export are supported.

[0030] The present application will be described in detail below with reference to specific embodiments: Example 1 The dental trauma prediction and information recording algorithm in this application is based on multimodal mechanical signals (acceleration, angular velocity, pressure) to achieve full-process data parsing. The specific implementation of each sub-algorithm module is as follows: Adaptive anti-interference data preprocessing algorithm: To address noise interference from saliva flow, chewing motion, and other factors in the oral cavity environment, an adaptive Kalman filter combined with sliding window smoothing preprocessing algorithm is designed to achieve accurate extraction of effective mechanical signals.

[0031] The adaptive Kalman filter implementation process includes the following steps: 1) Construction of state and observation equations: Define the state vector X = [a x ,a ,a z ,ω x ,ω ,ω z [,p,φ], where a x / a / a z For triaxial acceleration, ω x / ω / ω z Here, φ represents the triaxial angular velocity, p represents the pressure amplitude, and φ represents the signal phase; the observation equation is Z = HX + V, where H is the observation matrix and V is the observation noise.

[0032] Dynamic gain adjustment: By calculating the estimated value of the observation noise covariance matrix R in real time, the Kalman gain K is dynamically adjusted. When a signal change is detected (such as a potential impact signal), the gain is reduced to preserve the peak characteristics of the signal; when the signal is stable, the gain is increased to enhance the noise suppression effect.

[0033] 2) Iterative update: based on the predicted state value from the previous time step. k | k-1 Compared with the current observation value Z k Calculate the optimal estimate k = k | k-1 +K k (Z k -H k | k-1 And update the error covariance matrix P. k .

[0034] 3) Sliding Window Smoothing Optimization: A 50ms sliding window is used to process the filtered signal. A weighted average method is used to calculate the output value within the window. The weight of the data at the center of the window is set to 0.3, decreasing by 0.05 towards both sides to further reduce the interference of random noise on signal features and ensure the accuracy of subsequent feature extraction. Multi-Feature Fusion Direction and Force Prediction Algorithm: Based on the preprocessed effective signal, an improved Support Vector Machine (SVM) algorithm is used to simultaneously predict the direction and force of external injuries through multi-dimensional feature fusion.

[0035] 4) Feature extraction: Extracting peak value A from the acceleration signal p Peak time T p Rising slope K a Root mean square value R ams Four time-domain features; extracting the peak angular acceleration α from the angular velocity signal. p Two features are used to extract the pressure peak value P from the pressure signal: the integral value of angular displacement θ. p Pressure distribution entropy E p Two features are used to construct an 8-dimensional feature vector F=[A p ,T p ,K a ,R ams ,α p ,θ,P p E p ].

[0036] 5) Feature optimization: min-max normalization is used to map the feature vector to the [0,1] interval to eliminate the difference in dimensions; a weighted coefficient matrix W=diag[0.3,0.1,0.1,0.1,0.1,0.1,0.3,0.1] is introduced to give higher weight to the pressure peak and acceleration peak, and strengthen the influence of the core force characteristics.

[0037] 6) Improved SVM Model Construction: The RBF kernel function K(x,y)=exp(-||xy||² / (2σ²)) is adopted, and the kernel parameter σ (search range [0.01,10]) and the penalty factor C (search range [0.1,100]) are optimized by grid search method. The model is divided into a direction classification sub-model and a force regression sub-model. The direction classification sub-model calculates the cosine similarity between the feature vector and six direction templates (horizontal, vertical, and four oblique directions (45°, 135°, 225°, 315°)) and outputs the direction with the highest similarity as the prediction result. The force regression sub-model maps the feature vector to a specific force value F through SVM regression. p This allows for the quantification of intensity.

[0038] 7) Damage parameter identification algorithm based on decision fusion: When the predicted force exceeds the safety threshold, a fusion algorithm of decision tree + Bayesian classifier is adopted to achieve accurate identification of the degree of trauma and damage type. The identification standard refers to the guidelines of the International Association for Dental Trauma (IADT).

[0039] 8) Primary classification using decision trees: Constructing a multi-node decision tree, based on the strength value F. p , rate of change of force dF p / dt, force direction D, and pressure distribution range S are the decision nodes, with a threshold value of F. p ≤5N is considered a minor injury. <F p ≤20N is considered moderate trauma, F p >20N indicates severe trauma, completing the preliminary classification of trauma severity; at the same time, a preliminary set of candidate injury types is determined based on the direction and force characteristics (such as the candidate set of root fracture corresponding to a vertical high-force impact).

[0040] 9) Bayesian classifier optimization: The output of the decision tree is used as the prior probability P(Y|X1). Combined with the preset damage type-mechanical feature mapping library (constructed through massive clinical cases and simulation experiments), the posterior probability P(Y|X1,X2)∝P(X2|Y)P(Y|X1) is calculated, where X2 is the feature vector and Y is the damage type (coronal fracture, root fracture, subluxation, complete dislocation). The damage type with the highest posterior probability is output as the final result.

[0041] 10) Adaptive Threshold Correction: Based on the user's age (child / adult) and tooth type (deciduous teeth / permanent teeth), a linear correction function F is used to adjust the threshold. th '=F th ×k (k is a correction coefficient, k=0.7-0.9 for children, k=1.0 for adults; k=0.6-0.8 for deciduous teeth, k=1.0 for permanent teeth) dynamically adjusts the threshold of decision nodes to improve the algorithm's adaptability.

[0042] The principle behind this dental trauma prediction and information recording algorithm lies in constructing a processing procedure involving signal denoising, physical parameter analysis, and clinical decision mapping. Through multimodal signal fusion and adaptive computation, it achieves automatic analysis from raw oral biomechanical signals to precise damage assessment. Its working principle consists of the following three levels: First layer: Extracting real trauma mechanical signals in the oral cavity environment that is interfering with (saliva, chewing); Adaptive Kalman Filtering: Dynamically tracks signals using a state vector incorporating acceleration, angular velocity, pressure, and signal phase. Its key innovation lies in real-time noise estimation and gain (K value) adjustment: During signal abrupt changes (such as impacts): Automatically reduces gain to prioritize protecting the peak and rapidly changing characteristics of the impact signal, preventing them from being filtered out as noise. When the signal is stable: Increases gain to enhance the suppression of background noise. Sliding Window Smoothing: Performs a short-time-window (50ms) non-uniform weighted average on the filtered signal to further filter out residual high-frequency random noise. Simultaneously, through a central high-weight design, it reduces time-domain distortion of key signal features.

[0043] The second layer: quantitatively analyze the direction and force of the external injury from the purified signal.

[0044] Eight physically meaningful features (such as peak value, slope, and entropy) are extracted from triaxial acceleration, angular velocity, and pressure signals to form a multidimensional feature vector describing the impact event. A weighted matrix is ​​introduced, giving higher weights to the peak acceleration and peak pressure, enhancing the model's discriminative ability. An improved SVM dual-task model is used: employing template matching, it calculates the cosine similarity between the input features and six preset directional templates, outputting the best-matching direction. An optimized SVM regression model (RBF kernel) is used to nonlinearly map the high-dimensional feature vector to specific force values, achieving quantized output.

[0045] The third layer: Transforming physical parameters into medical diagnoses that conform to the International Association of Clinical Analyses (IADT).

[0046] Rapid initial screening using decision trees: Utilizing intuitive parameters such as force and direction as nodes, thresholds aligned with clinical experience (e.g., 5N, 20N) are set to rapidly and transparently screen for trauma severity grading and candidate injury types. Optimized decision using a Bayesian classifier: The output of the decision tree is used as a prior probability, combined with injury-feature statistical patterns (likelihood probability) constructed from massive case data. The posterior probability is calculated using the Bayesian formula, outputting the most probable injury type (e.g., crown fracture, root fracture), achieving an organic fusion of rule knowledge and data statistics, improving the robustness and accuracy of judgment. Adaptive threshold correction: The force threshold of the decision tree is dynamically adjusted based on the user's age (child / adult) and tooth type (deciduous / permanent), giving the algorithm personalized adaptability and better aligning with the physiological characteristics of different populations. All content not detailed in this application is prior art and can be implemented by those skilled in the art.

[0047] To implement the above algorithm, a dental trauma prediction and information recording system was constructed, with the following functions for each module: Wearable sensing module: Adopting a brace-like structure, it integrates a three-axis accelerometer (MPU-6050), a three-axis gyroscope, a miniature piezoresistive pressure sensor, and a capacitive contact trigger sensor to collect multimodal mechanical signals of the force on the teeth in the oral cavity, providing raw data input for the algorithm.

[0048] Algorithm Calculation Module: The STM32H743 microcontroller is used as the core, and the above-mentioned dental trauma prediction and information recording algorithm is used to complete the entire process of data preprocessing, direction and force prediction, and damage parameter identification.

[0049] Early warning module: Includes a built-in miniature vibrator and a portable audible and visual alarm. When the algorithm predicts a high-risk injury, it will remind the user to avoid the risk through dual warnings of vibration (100-300Hz) and audible and visual alarms (≥80dB).

[0050] Storage module: It adopts a dual architecture of local Flash (128MB) + cloud distributed storage, and encrypts the output of warning records (time, direction, intensity), trauma records (time, intensity, trauma degree, injury type) and raw sensor data.

[0051] Communication module: It adopts Bluetooth 5.2 + 5G dual-mode communication to realize real-time data transmission between the wearable sensing module and the computing module, and between the computing module and the terminal display module, ensuring rapid synchronization of algorithm results.

[0052] Terminal display module: including mobile APP and doctor workstation terminal, visually displaying various parameters output by the algorithm, supporting data query and export, and providing intuitive evidence of chief complaint for clinical diagnosis and treatment.

[0053] Example 2 This embodiment provides a method for predicting and recording information on dental trauma, including: (1) The wearable sensing module synchronously collects triaxial acceleration, triaxial angular velocity and local pressure signals. The wearable sensing module is rigidly coupled to the anterior tooth area, with an installation position error of ≤ ±0.3 mm and a sampling frequency of 10 kHz. The capacitive contact trigger signal is used as the starting mark of the impact event. (2) Based on the event flag, the adaptive Kalman filter is started, and the state vector represents the key physical quantities of the impact dynamics. The filter output enters a 50ms weighted sliding window for time-domain smoothing optimization, with the center having the highest weight and decreasing on both sides, taking into account both front-end fidelity and noise suppression. (3) Eight-dimensional time-domain features were extracted from the smoothed signal. The composition and weighting method were based on the damage mechanism defined by the International Association for Dental Trauma (IADT) guidelines. After normalization, standardized input was formed to support clinical interpretability. (4) An improved support vector machine model is used to jointly model the standardized features and output the impact direction category and the quantized force value (unit: N) simultaneously. The two share the feature space, loss function and training target to ensure physical consistency. (5) A rule-driven decision tree is constructed based on the IADT guidelines to screen the SVM output and generate a damage candidate set. Then, a Bayesian classifier is called, and the posterior probability of each damage type is calculated by combining the clinical prior knowledge base and likelihood modeling. (6) When the maximum posterior probability is ≥93.6% and the force is >20N and the direction is vertical or forward-upward oblique, send a command to the early warning execution unit to trigger a combined vibration and sound and light warning; (7) Generate a structured information record package containing a UTC nanosecond-level timestamp, three-dimensional force vector, impact direction, quantified force, ICD-10 encoded injury type, treatment recommendations, and original waveform SM3 hash value. The format meets the DICOM-RT extension specification and supports automatic parsing and electronic medical record writing by the hospital HIS system.

[0054] Specifically, the wearable sensing module is an adaptive structure for the anterior teeth region of the oral cavity. It can use a medical-grade silicone base with a thickness of 1.2mm and a hardness of Shore A35±2, meeting the biocompatibility requirements of ISO10993-5 and ASTM F2503 MRI compatibility requirements. This structure ensures patient safety and signal transmission fidelity. The silicone is vulcanized and molded, with a tear strength ≥12N / mm², and no leaching occurs after 72 hours of immersion. The hardness setting balances fit (too soft will result in slippage >0.8mm, too hard will compress the gums), with a measured installation error ≤±0.3mm. In practical applications, this module can be a removable occlusal pad, a lingual flexible patch, or a sensing inlay integrated into an invisible aligner.

[0055] The wearable sensing module completes sensor spatial calibration through a laser positioning mold and an optical measuring instrument. The installation position deviation of each sensing unit relative to the center of the labial surface of the tooth is ≤ ±0.3mm; the effective sensing area of ​​the pressure sensing unit is ≤4mm², matching the labial surface dimensions of a single central incisor (approximately 8mm wide × 10mm high). This calibration accuracy is an implicit premise of the IATT guidelines for impact point location—if the deviation is >0.5mm, a vertical impact will be misjudged as an oblique impact, increasing the rate of missed root fracture diagnosis by 22%. The laser mold uses 3D-printed photosensitive resin (Accura ClearVue) with a surface roughness Ra ≤ 0.8μm; the optical measuring instrument is a Keyence LK-H085 with a repeatability of ±0.1μm. In practical applications, the calibration method can be replaced by digital twin virtual assembly calibration.

[0056] The edge computing unit is an STM32H743 microcontroller running the Free RTOS real-time operating system. It has a 480MHz clock speed, supports floating-point operations and interrupt responses, with an acquisition latency of <1ms and a total processing time of ≤8ms for the entire filtering-feature-modeling process. This chip meets the requirements of GB / T 42061—2022, Clause 7.2, for medical device software execution platforms. At a 480MHz clock speed, it can process three channels of 10kHz raw data streams in parallel. Memory partitioning management includes a 256KB ADC cache, a 128KB filtering workspace, and a 64KB SVM inference area, all stored in ROM with no external dependencies.

[0057] The state vector of the adaptive Kalman filter contains key physical quantities characterizing impact dynamics, namely triaxial acceleration, triaxial angular velocity, local pressure amplitude, and signal phase angle, totaling eight dimensions, used to construct a clinically interpretable mechanical behavior model. This state dimension design directly corresponds to the formation mechanism of the six types of injuries in the IATT guidelines: such as a z Together with p, ω determines the risk of crown fracture; y Together with φ, it reflects the characteristics of rotational dislocation; φ (phase angle) is extracted through FFT phase spectrum and used to identify asymmetric impacts (such as unilateral impact vs. bilateral occlusion). The absence of any dimension will lead to a decrease of >15% in the model's ability to discriminate specific injury types. In practical applications, the phase angle can also be obtained by transient waveform zero-crossing detection.

[0058] The weighted sliding window has a length of 50ms, a center weight of 0.3, and adjacent sampling point weights of 0.25, 0.2, 0.15, and 0.1, respectively, meeting the event window fidelity requirements of Clause 5.4 of GB / T 42061—2022. This weight distribution has been validated through clinical impact simulations and real cases, effectively balancing artifact suppression and peak shape preservation. Compared to triangular windows (of equal length) and rectangular windows, this scheme reduces the false trigger rate to 1.2% under chewing interference, while the traditional rectangular window reaches 18.5%. In practical applications, the weights can be automatically optimized according to the objective function (e.g., minimizing Ep fluctuations).

[0059] The 8-dimensional features include: peak acceleration Peak time Rising slope Root mean square Peak angular acceleration Angular displacement integral Peak pressure Pressure distribution entropy Each dimension originates from the quantitative description of the injury mechanism in the IATT guidelines. The 8-dimensional features are normalized to eliminate dimensional differences; the normalization method is at least one of min-max, Z-score, or Robust Scaling, where the upper and lower limits of the min-max normalization are derived from the statistical boundaries of the IATT clinical database (covering 99.9% of the samples). Normalization is a prerequisite for SVM convergence and must be based on the clinical data distribution. For example, , The P99.9 quantile from 1268 labeled data; the Z-score method can be used for outlier robust scenarios (such as abnormal movements like epileptic seizures).

[0060] The improved support vector machine model uses the RBF kernel function, with the kernel width parameter σ determined to be 0.42 through grid search. The penalty factor C=25.6, and the training, validation, and test sets are split in a 6:2:2 ratio, meeting the clinical model development guidelines recommended by the IATT guidelines. σ=0.42 achieves a maximum margin of 0.85 (normalized space) for directional classification boundaries, significantly better than σ=0.1 (overfitting) or σ=1.0 (underseparation). C=25.6 was determined through five-fold cross-validation to balance preventing overfitting and maintaining shock sensitivity. The dataset is stratified by region, age, and gender to ensure generalization ability.

[0061] The warning execution unit includes a vibration motor, a red and blue dual-color LED array, and a buzzer. Its action command is issued by the edge computing unit when the force is greater than 20N and the direction is vertical or forward-upward oblique. The vibration frequency is 250Hz, lasts for 1.5s, and the amplitude is 0.2g. The LED brightness is ≥300cd / m², and the buzzer sound pressure level is 85dB@10cm, all complying with the clinical warning performance standards of IEC 60601-1-6:2022, Clause 8.3. This unit is a clinical-grade safety component, not a general-purpose reminder device. The vibration frequency of 250Hz is within the human tactile sensitive zone (200–300Hz), ensuring a wake-up effect. The LED uses a combination of red (625nm) and blue (470nm) to avoid the enamel excitation band (<405nm), preventing accidental activation of dentinal tubules. The buzzer uses pulse modulation (30% duty cycle) to avoid auditory fatigue caused by continuous sound pressure. In practical applications, it can be expanded into tongue microcurrent feedback or AR glasses overlay prompts.

[0062] The structured information record package is in JSON Schema v1.1 format, and the fields include UTC nanosecond timestamp, three-dimensional force vector, impact direction, quantized force (N), ICD-10 encoding (such as K03.4), standardized treatment recommendations, and original waveform SM3 hash value; it supports DICOM-RT extended fields, can be parsed by the hospital PACS system and automatically associated with CBCT images.

[0063] Example 3 The third objective of this application is to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned method for predicting and recording dental trauma.

[0064] The method for predicting and recording dental trauma implemented by the processor when executing a computer program specifically includes: S1. Signal Acquisition: The wearable sensing module acquires raw signals of triaxial acceleration, triaxial angular velocity, and multimodal mechanical stress on teeth in the oral cavity. S2. Adaptive anti-interference data preprocessing: The original signal is processed by a combination algorithm of adaptive Kalman filtering and sliding window smoothing to extract the effective mechanical signal; S3. Direction and force prediction of multi-feature fusion: Based on the preprocessed effective signal, multi-dimensional features are extracted and optimized, and the force direction and force of dental trauma are predicted simultaneously through an improved support vector machine algorithm. S4. Damage parameter identification through decision fusion: If the predicted force exceeds the safety threshold, a fusion algorithm of decision tree + Bayesian classifier is used, combined with adaptive threshold correction rules, to identify the degree of trauma and damage type. The identification criteria refer to the International Association for Dental Trauma (IADT) guidelines. S5. Warning Trigger: If the injury is determined to be of high risk, a dual warning will be triggered through the warning module.

[0065] Example 4 The fourth objective of this application is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a method for predicting and recording dental trauma.

[0066] When a computer program is executed by a processor, the method for predicting and recording dental trauma includes: S1. Signal Acquisition: The wearable sensing module acquires raw signals of triaxial acceleration, triaxial angular velocity, and multimodal mechanical stress on teeth in the oral cavity. S2. Adaptive anti-interference data preprocessing: The original signal is processed by a combination algorithm of adaptive Kalman filtering and sliding window smoothing to extract the effective mechanical signal; S3. Direction and force prediction of multi-feature fusion: Based on the preprocessed effective signal, multi-dimensional features are extracted and optimized, and the force direction and force of dental trauma are predicted simultaneously through an improved support vector machine algorithm. S4. Damage parameter identification through decision fusion: If the predicted force exceeds the safety threshold, a fusion algorithm of decision tree + Bayesian classifier is used, combined with adaptive threshold correction rules, to identify the degree of trauma and damage type. The identification criteria refer to the International Association for Dental Trauma (IADT) guidelines. S5. Warning Trigger: If the injury is determined to be of high risk, a dual warning will be triggered through the warning module.

[0067] Example 5 The fifth objective of this application is to provide a computer program product, which includes computer instructions that instruct a computer to perform a method for predicting and recording dental trauma.

[0068] S1. Signal Acquisition: The wearable sensing module acquires raw signals of triaxial acceleration, triaxial angular velocity, and multimodal mechanical stress on teeth in the oral cavity. S2. Adaptive anti-interference data preprocessing: The original signal is processed by a combination algorithm of adaptive Kalman filtering and sliding window smoothing to extract the effective mechanical signal; S3. Direction and force prediction of multi-feature fusion: Based on the preprocessed effective signal, multi-dimensional features are extracted and optimized, and the force direction and force of dental trauma are predicted simultaneously through an improved support vector machine algorithm. S4. Damage parameter identification through decision fusion: If the predicted force exceeds the safety threshold, a fusion algorithm of decision tree + Bayesian classifier is used, combined with adaptive threshold correction rules, to identify the degree of trauma and damage type. The identification criteria refer to the International Association for Dental Trauma (IADT) guidelines. S5. Warning Trigger: If the injury is determined to be of high risk, a dual warning will be triggered through the warning module.

[0069] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0070] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0071] This application may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application may take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, readable storage media, optical storage, etc.) containing computer-usable program code.

[0072] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0073] Obviously, the described embodiments are only some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort should fall within the scope of protection of this application.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A system for predicting and recording dental trauma, characterized in that, It includes a wearable sensing module and an algorithm processing module that are connected in sequence. The algorithm processing module communicates bidirectionally with an early warning module, a storage module, and a communication module, respectively. The communication module communicates bidirectionally with a terminal display module. The wearable sensing module has a brace-like structure and integrates a three-axis accelerometer, a three-axis gyroscope, a miniature piezoresistive pressure sensor, and a capacitive contact trigger sensor to collect multimodal mechanical signals of force on teeth in the oral cavity. The algorithm operation module is based on a microcontroller and has an algorithm for predicting dental trauma and recording information. It completes the entire process of data preprocessing, direction and force prediction, and damage parameter identification. The early warning module includes a built-in miniature vibrator in the braces and a portable audible and visual alarm, providing dual early warning for high-risk injuries.

2. The dental trauma prediction and information recording system according to claim 1, characterized in that, Also includes: The storage module adopts a dual architecture of local and cloud distributed storage for encrypted storage of dental trauma-related data; The communication module adopts dual-mode communication to realize real-time data transmission between modules; The terminal display module is used to visually display the algorithm output parameters and support data query and export.

3. The dental trauma prediction and information recording system according to claim 1, characterized in that, The encrypted data stored in the storage module includes early warning records, trauma records, and raw sensor data; among them, the early warning records include time, force direction, and predicted force, and the trauma records include time, predicted force, degree of trauma, and type of injury.

4. The dental trauma prediction and information recording system according to claim 1, characterized in that, The wearable sensing module has a three-axis accelerometer, specifically an MPU-6050. In the early warning module, the vibration frequency of the micro vibrator is 100-300Hz, and the sound pressure level of the portable audible and visual alarm is ≥80dB.

5. A method for predicting and recording dental trauma, applied to the dental trauma prediction and recording system as described in any one of claims 1 to 4, characterized in that, Includes the following steps: S1. Collect raw signals of triaxial acceleration, triaxial angular velocity, and multimodal mechanical stress on teeth in the oral cavity through a wearable sensing module; S2. Adaptive Kalman filtering and sliding window smoothing algorithms are used to process the original signal and extract the effective mechanical signal; S3. Based on the preprocessed effective signal, extract multi-dimensional features and optimize them. Simultaneously predict the force direction and intensity of dental trauma using an improved support vector machine algorithm. S4. If the predicted intensity exceeds the safety threshold, a fusion algorithm of decision tree and Bayesian classifier is used, combined with adaptive threshold correction rules, to identify the degree of trauma and injury type. S5. If the injury is determined to be of high risk, an early warning will be activated through the early warning module.

6. The method for predicting and recording dental trauma according to claim 5, characterized in that, The implementation process of the adaptive Kalman filter described in S2 includes: Define the state vector X=[a x ,a ,a z ,ω x ,ω ,ω z [,p,φ], where a x / a / a z For triaxial acceleration, ω x / ω / ω z Here, φ represents the triaxial angular velocity, p represents the pressure amplitude, and φ represents the signal phase; the observation equation is Z = HX + V, where H is the observation matrix and V is the observation noise. By calculating the estimated value of the observation noise covariance matrix R in real time, the Kalman gain K is dynamically adjusted. When the signal changes abruptly, the gain is reduced to preserve peak characteristics, and when the signal is stable, the gain is increased to enhance noise suppression. Based on the state prediction value at the previous moment k | k-1 Compared with the current observation value Z k Calculate the optimal estimate k = k | k-1 +K k (Z k -H k | k-1 And update the error covariance matrix P. k .

7. The method for predicting and recording dental trauma according to claim 5, characterized in that, The specific process of smoothing the sliding window described in S2 is as follows: A sliding window with a length in milliseconds is used to process the filtered signal. The output value is calculated using a weighted average method within the window. The weight of the data in the center of the window is set to 0.3, and the weight of the data on both sides decreases by 0.

05.

8. The method for predicting and recording dental trauma according to claim 5, characterized in that, The extraction process of the multi-dimensional features is as follows: Extracting peak value A from acceleration signal p Peak time T p Rising slope K a Root mean square value R am Temporal characteristics; extraction of peak angular acceleration α from angular velocity signal p Two characteristics: angular displacement integral value θ; Extracting the pressure peak P from the pressure signal p Pressure distribution entropy E p Construct an 8-dimensional feature vector F=[A p ,T p ,K a ,R ams ,α p ,θ,P p E p ]; The process of extracting and optimizing multi-dimensional features based on the preprocessed effective signal is as follows: First, min-max normalization is used to map the 8-dimensional feature vector to the [0,1] interval to eliminate the difference in dimensions; then, a weighted coefficient matrix W=diag[0.3,0.1,0.1,0.1,0.1,0.1,0.3,0.1] is introduced to weight the feature vector, thereby strengthening the influence of the core force characteristics of the peak acceleration and peak pressure.

9. The method for predicting and recording dental trauma according to claim 5, characterized in that, The construction process of the improved support vector machine algorithm is as follows: The kernel function K(x,y)=exp(-||xy||² / (2σ²)) is adopted, and the kernel parameter σ and the penalty factor C are optimized by grid search method, where the search range of σ is [0.01,10] and the search range of C is [0.1,100]. The direction classification sub-model calculates the cosine similarity between the feature vector and six direction templates (horizontal, vertical, 45°, 135°, 225°, and 315°), and outputs the direction with the highest similarity as the prediction result. The force regression sub-model maps the feature vector to a specific force value F through SVM regression. p This allows for the quantification of intensity.

10. The method for predicting and recording dental trauma according to claim 5, characterized in that, The initial classification process of the decision tree is as follows: using the strength value F p , rate of change of force dF p / dt, force direction D, and pressure distribution range S are the decision nodes, and the force threshold F is set. p ≤5N is considered a minor injury. <F p ≤20N is considered moderate trauma, F p >20N indicates severe trauma, completing the preliminary classification of trauma severity; simultaneously, based on the direction and intensity of force, a preliminary candidate set of injury types is determined; The optimization process of the Bayesian classifier is as follows: the output of the decision tree is used as the prior probability P(Y|X1), and combined with the preset damage type-mechanical feature mapping library, the posterior probability P(Y|X1,X2)∝P(X2|Y)P(Y|X1) is calculated, where X2 is the optimized 8-dimensional feature vector and Y is the damage type; the damage type with the highest posterior probability is output as the final result; the damage type-mechanical feature mapping library is constructed through massive clinical cases and simulation experiments, and the damage types include coronal fracture, root fracture, subluxation, and complete dislocation; The specific rule for the adaptive threshold correction is as follows: based on the user's age and tooth type, a linear correction function F is used. th '=F th ×k dynamically adjusts the threshold of the decision node, F th F is the original threshold. th ' is the corrected threshold, and k is the correction coefficient; where, for children k=0.7-0.9, for adults k=1.0; for deciduous teeth k=0.6-0.8, for permanent teeth k=1.0.

Images