AI enhanced real-time occlusion detection system based on intraoral image

Abstract

The invention relates to the field of oral medical diagnosis, and discloses an intraoral image-based AI enhanced real-time occlusion detection system, which comprises a multi-source intraoral image acquisition module for acquiring dynamic oral image data through a high-definition oral endoscope and recording a three-dimensional structure of a tooth surface by combining a laser scanning device, meanwhile, force distribution of an occlusion contact point is collected through a pressure sensor array; the multi-source intraoral image acquisition module comprises a timestamp synchronization unit and is used for performing multi-source synchronization on an oral image frame sequence and occlusal force distribution data by utilizing a unified timestamp; through the multi-source in-port image acquisition module, synchronous acquisition of dynamic images, three-dimensional structure data and bite force distribution is realized, and time sequence consistency of the data is ensured through timestamp marking. Compared with a traditional single data source collection mode, the multi-source collection can comprehensively record dynamic characteristics in the occlusion process, rich and accurate input is provided for subsequent AI analysis, and data missing or collection deviation is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of oral medical diagnosis, and in particular to an AI-enhanced real-time occlusion detection system based on intraoral images. Background Art

[0002] Oral health is one of the important factors affecting human overall health. Among them, abnormal occlusal relationship is closely related to many diseases such as temporomandibular joint disease, bruxism, and chronic oral pain. Clinically, occlusal detection and analysis are of great significance for predicting temporomandibular joint disorders, formulating orthodontic treatment plans, and evaluating the effects of orthodontic treatment. However, occlusal detection usually requires a combination of dynamic images, three-dimensional structural data, mechanical distribution and other information, which places high demands on technology and equipment. At the same time, traditional occlusal detection methods often rely on the doctor's experience and judgment, and lack objective, real-time and high-precision analysis tools, which greatly limits the diagnostic efficiency and treatment quality of oral medical care.

[0003] In the prior art, the equipment and methods for occlusion detection mainly include digital pressure detection systems and oral 3D scanners, which can provide pressure distribution information and tooth surface geometry information when the patient is biting. However, these systems usually have a single function and can only collect a certain type of data, making it difficult to fully record the dynamic changes during the occlusion process. In addition, although some existing intelligent detection systems integrate multiple data sources, the collection and analysis process of multi-source data often has problems such as time series asynchrony and information isolation, making it impossible to achieve real-time, high-precision comprehensive diagnosis. At the same time, the real-time performance and deployment flexibility of existing systems are poor, making it difficult to adapt to scenarios with limited resources.

[0004] There is a key problem in the existing technology: it is impossible to achieve efficient collection and real-time fusion of multi-source data, and it is difficult to provide comprehensive and accurate occlusal detection and diagnosis results. This problem is particularly obvious when the current multimodal data analysis technology is underdeveloped and the multi-device collaboration capability is limited. In order to solve this problem, a system is needed that can achieve synchronous collection, efficient fusion and intelligent diagnosis of multi-source data, so as to provide real-time and accurate occlusal detection support for clinical practice. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides an AI-enhanced real-time occlusion detection system based on intraoral images, which solves the problem of being unable to achieve efficient collection and real-time fusion of multi-source data and difficulty in providing comprehensive and accurate occlusion detection and diagnosis results.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: an AI-enhanced real-time occlusal detection system based on intraoral images, including a multi-source intraoral image acquisition module, which is used to obtain dynamic oral image data through a high-definition oral endoscope, and record the three-dimensional structure of the tooth surface in combination with a laser scanning device, and at the same time collect the force distribution of the occlusal contact point through a pressure sensor array;

[0007] The multi-source intraoral image acquisition module includes a time stamp synchronization unit, which uses a unified time stamp to perform multi-source synchronization on the oral image frame sequence and the bite force distribution data;

[0008] A data preprocessing and enhancement module, connected to the multi-source intraoral image acquisition module, is used to perform noise filtering, illumination balance and geometric distortion calibration on the acquired image data, and generate multi-angle samples through data enhancement methods such as random rotation, cropping and mirror flipping;

[0009] An AI-based occlusion detection module, connected to the data preprocessing and enhancement module, is used to perform feature extraction, occlusion point detection, and dynamic change modeling on the processed data through a deep learning model;

[0010] A real-time dynamic feedback and abnormality analysis module, connected to the AI-based occlusion detection module, is used to generate an occlusion point distribution map, a force distribution heat map, and analyze the occlusion data;

[0011] A multimodal data fusion module, connected to the AI-based occlusion detection module and the real-time dynamic feedback and abnormality analysis module, respectively, for fusing image data, occlusal force data and patient medical record information;

[0012] The lightweight edge computing and system deployment module is connected to the above modules to optimize the computing efficiency of AI models and support collaborative computing between edge devices and the cloud.

[0013] Preferably, a high-definition oral endoscope is used to collect dynamic image data with a resolution greater than 2K and a frame rate ≥ 60fps;

[0014] Laser scanning equipment to capture the three-dimensional structure and surface texture of teeth;

[0015] An array of micro pressure sensors to collect force distribution at the occlusal contact points;

[0016] The timestamp synchronization unit is used to add unified time stamps to dynamic images and force distribution data.

[0017] Preferably, a noise reduction processing unit is connected to the multi-source intraoral image acquisition module and is used to reduce noise on the image data based on non-local mean filtering technology and spectral domain de-reflection technology;

[0018] An optical correction unit for adjusting the brightness and contrast of the image based on an adaptive illumination equalization algorithm;

[0019] A geometric calibration unit, used to correct the geometric distortion of the image through a feature point matching algorithm;

[0020] The data enhancement unit is used to randomly rotate, crop and mirror-flip the rectified image data to generate diversified image samples.

[0021] Preferably, a deep learning model is connected to the data preprocessing and enhancement module, and is used to extract multi-scale features through a feature pyramid network;

[0022] Attention mechanism module, embedding convolutional block attention module to enhance the detection ability of small objects;

[0023] Dynamic modeling unit, used for temporal modeling of image frame sequences based on bidirectional LSTM and temporal convolutional networks;

[0024] The classification and localization unit is used to optimize the category distribution of bite point detection based on focal loss, and optimize the localization of the target bounding box through IoU loss.

[0025] Preferably, a transfer learning module is used to perform pre-training on a general object detection dataset and adapt to oral imaging data through fine-tuning;

[0026] Adaptive hyperparameter optimization module, which is used to adjust the learning rate, batch size, and regularization parameters through the Bayesian optimization algorithm.

[0027] Preferably, a real-time feedback unit is connected to the AI-based occlusion detection module to generate an occlusion point distribution map and a force distribution heat map;

[0028] An abnormality analysis unit, used to extract occlusal abnormality features from the test results and generate a diagnosis report;

[0029] Visual interface unit, used to dynamically display three-dimensional occlusal data and compare and analyze historical data.

[0030] Preferably, a data alignment unit is connected to the AI-based occlusion detection module and is used to uniformly encode the image data and mechanical data through a coordinate channel attention mechanism;

[0031] Joint modeling unit, used to interactively model imaging data, mechanical characteristics, and medical record data based on the Transformer framework;

[0032] Fusion analysis unit for generating comprehensive occlusal diagnosis results and correction recommendations.

[0033] Preferably, a model compression unit is connected to the AI-based occlusion detection module and is used to compress the model through pruning and quantization techniques;

[0034] Knowledge distillation module, used to optimize the lightweight student model through the teacher model;

[0035] The edge-cloud collaboration unit is used to complete real-time reasoning on edge devices and migrate complex computing tasks to the cloud.

[0036] Preferably, the detection function of the system completes the real-time detection of dynamic bite points and contact points by connecting a multi-source intraoral image acquisition module, a data preprocessing and enhancement module, and an AI-based occlusion detection module.

[0037] Preferably, the system is connected to the hospital information system via an API interface to achieve sharing and synchronization of occlusal detection data and support cross-device diagnostic record updates.

[0038] The present invention provides an AI-enhanced real-time occlusion detection system based on intraoral images.

[0039] It has the following beneficial effects:

[0040] 1. The present invention realizes the synchronous acquisition of dynamic images, three-dimensional structure data and occlusal force distribution through a multi-source intraoral image acquisition module, and ensures the temporal consistency of the data through timestamp marking. Compared with the traditional single data source acquisition method, multi-source acquisition can comprehensively record the dynamic characteristics of the occlusal process, provide rich and accurate input for subsequent AI analysis, and avoid data omission or acquisition deviation.

[0041] 2. The present invention uses data preprocessing and enhancement modules to perform multiple optimization processes such as noise reduction, illumination balance, and geometric correction on the collected data, and uses data enhancement technology to expand the diversity of samples. These technologies greatly improve the quality of input data and the generalization ability of the model, enabling the system to accurately identify bite points and contact points in complex oral environments, and improving the robustness of the system in different scenarios.

[0042] 3. The AI-based occlusion detection module of the present invention uses the YOLOv8 deep learning model combined with dynamic modeling technology to perform high-precision detection of occlusion points and contact points, and captures the time series change characteristics of the occlusion process through dynamic modeling. Compared with traditional manual analysis, this module can quickly and accurately identify occlusion abnormalities and provide reliable data support for real-time diagnosis.

[0043] 4. The real-time dynamic feedback and abnormal analysis module of the present invention realizes the dynamic display of the occlusal process by generating the occlusal point distribution map, force distribution heat map and dynamic change curve, and combines the rule analysis with the AI ​​detection results to identify problems such as eccentric bite, overload and contact abnormality. This intuitive feedback method not only improves the doctor's diagnostic efficiency, but also provides patients with a visual occlusal health assessment.

[0044] 5. The multimodal data fusion module of the present invention deeply integrates dynamic images, force distribution and medical record data through joint modeling to generate comprehensive diagnostic results, including occlusal abnormality scores, abnormal pattern classification and diagnostic suggestions. Compared with single data source analysis, multimodal fusion can capture deeper correlation features, provide doctors with more comprehensive diagnostic basis, and improve the intelligence level of the system.

[0045] 6. The present invention uses lightweight edge computing and system deployment modules. The present invention uses model compression and hardware acceleration technology to achieve real-time reasoning on embedded devices, and distributes complex tasks to the cloud for processing through edge-cloud collaboration. This design not only ensures the real-time and stability of the system, but also allows efficient operation in resource-constrained devices, and is suitable for clinical and mobile medical scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is the main framework diagram of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] Please see attached Figure 1 The embodiment of the present invention provides an AI-enhanced real-time occlusion detection system based on intraoral images, comprising:

[0049] Multi-source intraoral image acquisition module

[0050] For the multi-source intraoral image acquisition module, the core goal of the system is to achieve comprehensive data collection of the patient's occlusal dynamics through the collaborative work of multiple hardware devices, including core information such as dynamic oral images, three-dimensional tooth structure and bite force distribution. The collection of multi-source data is the basis for subsequent processing and AI analysis of the system. At the same time, the time synchronization mechanism is used to ensure the timing consistency between different data sources. The design of the multi-source intraoral image acquisition module needs to consider many factors, including data acquisition accuracy, synchronization requirements and hardware reliability, in order to meet the needs of real-time occlusion detection.

[0051] Module composition and function

[0052] In this embodiment, the multi-source intraoral image acquisition module includes the following components:

[0053] High-definition oral endoscope: used to collect high-resolution image data of the patient's bite dynamics;

[0054] Laser scanning equipment: used to record the three-dimensional structural features of the patient's tooth surface;

[0055] Pressure sensor array: used to obtain real-time force distribution data of the occlusal contact point;

[0056] Timestamp synchronization unit: used to add unified time stamps to the collected dynamic image data, three-dimensional structure data and mechanical data.

[0057] Data collection process

[0058] In this embodiment, the working steps of the multi-source intraoral image acquisition module are as follows:

[0059] The high-definition oral endoscope collects dynamic images of the patient's oral cavity in real time, with a resolution of 2K or above and a frame rate of 60fps. The image data is mainly used to capture the dynamic changes in the patient's bite process, including the formation and disappearance of tooth contact points. The endoscope is designed to be handheld, which makes it easier for doctors to locate key areas of the patient's oral cavity.

[0060] The laser scanning device acquires the three-dimensional geometric shape of the patient's teeth through a non-contact laser scanning method. The scanning resolution of the device reaches 50μm, which is used to record the detailed features of the tooth surface, such as the position of the bite point and its surrounding geometric relationship. During the three-dimensional scanning process, the device automatically calibrates the laser angle and focal length to ensure scanning accuracy.

[0061] The pressure sensor array is arranged on the occlusal contact surface of the patient's teeth to collect the force distribution of the occlusal contact point. The single-point detection accuracy of each sensor is 1N, and the sampling frequency is 500Hz. The pressure sensor sends the real-time collected data to the processing module through the data transmission unit to ensure that the mechanical information can be updated synchronously.

[0062] The timestamp synchronization unit uses a unified time base to synchronize the three types of data (dynamic images, three-dimensional structure and force distribution) to ensure the temporal consistency of multi-source data. The timestamp information is recorded in microseconds to ensure that the data can be accurately matched to the bite action at the same time point in subsequent processing.

[0063] High Definition Intraoral Camera:

[0064] In this embodiment, the endoscope uses a multi-layer anti-reflective coating lens to improve imaging clarity and integrates a high-sensitivity CMOS sensor for high-definition imaging in low-light environments. The endoscope has a built-in ring LED light source to provide uniform lighting and eliminate image interference caused by local shadows. The video output of the endoscope is connected to the pre-processing module via a high-speed data interface (such as USB 3.0) to support real-time data transmission.

[0065] Laser scanning equipment:

[0066] In this embodiment, the laser scanning device uses point cloud scanning technology to generate a complete three-dimensional structure of the tooth through multiple scans. The device has a built-in adaptive focal length adjustment system to adjust the laser focus position according to the depth of the patient's teeth to avoid data loss or overlap. The laser scanning data is compressed through point cloud coding and transmitted to the data processing module to reduce the transmission bandwidth occupation.

[0067] Pressure sensor array:

[0068] In this embodiment, the sensor array is made of piezoelectric ceramic material, which has high sensitivity and fast response capability. The arrangement of the sensor array is adjusted according to the actual occlusal morphology of the patient's teeth to ensure that the force distribution data covers all contact areas. The sensor data is synchronously transmitted through a multi-channel data acquisition unit with a sampling frequency of 500Hz, supporting real-time monitoring of force distribution changes.

[0069] Timestamp synchronization unit:

[0070] In this embodiment, the timestamp synchronization unit provides a unified time reference through a high-precision crystal oscillator, and time-aligns the data streams of the endoscope, laser scanning device, and pressure sensor through a synchronization protocol. The timestamp is recorded with microsecond accuracy, and each frame of image data, each scanning point, and each set of force distribution data is accompanied by a timestamp to ensure data synchronization.

[0071] Data Output

[0072] In this embodiment, the output data of the multi-source intraoral image acquisition module includes:

[0073] Dynamic image data: records the patient's bite dynamics at a frequency of 60 frames per second in the form of a high-resolution video stream;

[0074] 3D structure data: Output the 3D geometric information of the tooth surface in point cloud data format with a data storage accuracy of 50μm;

[0075] Occlusal force distribution data: Output the force value of each occlusal contact point in the form of a force matrix, in Newton (N);

[0076] Timestamp information: A timestamp is added to each set of data to mark the time of collection.

[0077] In this embodiment, the output data of the multi-source intraoral image acquisition module is directly transmitted to the data preprocessing and enhancement module. The high-definition dynamic image data is used for image noise reduction and geometric calibration; the three-dimensional structure data is used for point cloud optimization; and the bite force distribution data is used to generate pressure mapping. The timestamp synchronization information is used to time-align the data stream in the preprocessing stage to ensure that the subsequent processing module can accurately process the multi-source data.

[0078] Data preprocessing and enhancement module

[0079] The main function of the data preprocessing and enhancement module is to standardize the data output by the multi-source intraoral image acquisition module, including noise reduction, illumination balance, geometric correction and data enhancement, to optimize data quality and provide high-quality input data for subsequent AI detection. The design of this module needs to combine the characteristics of multi-source data and downstream analysis requirements to ensure that the processed data meets the accuracy requirements of occlusal detection in terms of spatial consistency, temporal continuity and content integrity.

[0080] Image preprocessing: mainly includes noise filtering, illumination equalization and geometric distortion calibration. These processes are to improve image quality, reduce unnecessary noise, improve the accuracy of model training, and avoid overfitting.

[0081] Data enhancement: including random rotation, cropping, mirror flipping and other methods, aims to expand the training set through diversified training data and increase the generalization ability of the model.

[0082] In this embodiment, the data preprocessing and enhancement module receives the dynamic image data, three-dimensional structure data and force distribution data from the multi-source intraoral image acquisition module, and performs corresponding processing on these data, including:

[0083] Noise reduction, illumination equalization and geometric correction of image data;

[0084] Point cloud optimization and data compression of 3D structure data;

[0085] Normalization of force distribution data;

[0086] Data enhancement technology generates samples with multiple angles and features.

[0087] The goal of this module is to improve the quality and adaptability of data through a series of normalization and enhancement techniques, providing more accurate input data for AI models.

[0088] Noise reduction

[0089] In this embodiment, a non-local mean filtering algorithm (NLM) is used to eliminate random noise in dynamic images. The algorithm removes random noise while retaining key details by analyzing the similarity of adjacent pixels. In addition, for optical artifacts (such as reflected light spots) in dynamic images, spectral domain de-reflection technology is used to locate and remove abnormal high-frequency reflection signals by analyzing the image spectral frequency distribution, thereby optimizing the visual quality of the image.

[0090] Balanced lighting

[0091] In this embodiment, in order to solve the local brightness difference caused by uneven illumination in dynamic images, an adaptive illumination balancing algorithm is used. The specific method is as follows:

[0092] By calculating the local histogram of the image, the brightness range of the pixel distribution is obtained;

[0093] Apply histogram equalization technology to optimize the contrast of local areas;

[0094] Gaussian smoothing operation is introduced to prevent over-sharpening of images after illumination adjustment.

[0095] This technology evens out the overall brightness of the image while preserving the texture and morphological details of the tooth surface.

[0096] Geometric correction

[0097] In this embodiment, in order to process geometric distortion caused by imaging angle or device jitter, a method combining SIFT (Scale-Invariant Feature Transform) algorithm and RANSAC (Random Sample Consensus) algorithm is adopted:

[0098] Use SIFT algorithm to extract key feature points in the image;

[0099] The robustness of the feature point matching results is optimized through the RANSAC algorithm to remove the incorrectly matched feature points;

[0100] A geometric correction model is generated based on the matching results, and the overall geometric adjustment of the image is performed to maintain spatial consistency.

[0101] Geometric correction ensures the precise correspondence between each frame of the dynamic image and the actual tooth structure.

[0102] Data Augmentation

[0103] In this embodiment, in order to improve the AI ​​model's ability to learn diverse occlusal features, the following data enhancement techniques are used:

[0104] Random rotation: The dynamic image is randomly rotated at a certain angle (±15°) in the vertical direction to simulate the possible slight movement of the patient's head when biting;

[0105] Mirror flip: Mirror the image data horizontally to expand sample diversity;

[0106] Random cropping and scaling: Randomly select local areas of the image for cropping and adjust the scaling ratio to retain the key occlusal areas;

[0107] Generative Adversarial Network (GAN): Based on the collected original image data, GAN technology is used to generate simulated images in multi-angle and multi-lighting environments to further expand the data set.

[0108] Processing of 3D structure data

[0109] In this embodiment, the following optimization is performed on the three-dimensional point cloud data output by the laser scanning device:

[0110] Point cloud denoising: Remove outliers through a neighborhood density-based algorithm to optimize the integrity of 3D data;

[0111] Data compression: Octree Encoding technology is used to compress point cloud data into voxel representation, reducing transmission bandwidth and storage requirements.

[0112] Processing of force distribution data

[0113] In this embodiment, the force distribution data collected by the pressure sensor array is processed as follows:

[0114] Data normalization: According to the maximum and minimum force ranges, the force distribution value is mapped to the range of 0-1 to facilitate subsequent AI model processing;

[0115] Time series smoothing: The weighted moving average algorithm is used to perform time series smoothing on the force distribution data to remove instantaneous noise in mechanical sampling.

[0116] Data Output

[0117] In this embodiment, the output data of the data preprocessing and enhancement module includes:

[0118] Dynamic image data after noise reduction, illumination balance and geometric correction;

[0119] The optimized 3D point cloud data is output in a compressed format;

[0120] Normalized force distribution data is output in matrix form.

[0121] All data are time-stamped before output to ensure synchronization of multi-source data and occlusion time.

[0122] In this embodiment, the data preprocessing and enhancement module receives the raw data output by the multi-source intraoral image acquisition module, and after completing the preprocessing, transmits the processed image data, three-dimensional structure data and force distribution data to the AI-based occlusion detection module. This module not only ensures the synchronization of multi-source data through timestamps, but also expands the training samples of the AI ​​model through data enhancement technology, providing high-quality input for subsequent occlusal point detection and mechanical analysis.

[0123] AI-based occlusion detection module

[0124] The main task of the AI-based occlusion detection module is to receive pre-processed multi-source data (dynamic images, three-dimensional structure data, force distribution data), detect the occlusion points and contact points through deep learning algorithms, analyze the characteristics of dynamic changes during the occlusion process, and identify abnormal situations. This module is the core analysis module of the system, relying on artificial intelligence technology to achieve high-precision, real-time occlusion feature extraction, and provide basic data support for subsequent feedback and fusion modules.

[0125] In this embodiment, the AI-based occlusion detection module uses deep learning technology to perform feature extraction, detection and dynamic modeling on multi-source data. The specific functions include:

[0126] Multi-scale extraction of dynamic image features;

[0127] Location and classification of bite and contact points;

[0128] Time series modeling during the occlusion process to extract dynamic change features;

[0129] In the dynamic change modeling part, the main goal is to process time series data in order to capture the long-term and short-term dependencies of bite points and dynamic changes. To achieve this goal, LSTM (Long Short-Term Memory Network) is used, which is a deep learning model widely used in time series modeling and sequence data prediction. Its core advantage is that it can capture long-term dependency information in time series and overcome the gradient vanishing problem of traditional neural networks when dealing with long-term time dependency problems.

[0130] LSTM network model architecture

[0131] LSTM is a variant based on recurrent neural network (RNN), which not only processes the current input data at each time step, but also keeps the memory of past information. It consists of three important gating mechanisms: forget gate, input gate and output gate, which are used to control the forgetting, storage and output of information respectively.

[0132] Forget gate: controls how much historical information is forgotten from the memory unit.

[0133] Input gate: determines how much new information to store into the memory cell.

[0134] Output gate: determines what information in the memory cell the current output should be based on.

[0135] This mechanism enables LSTM to effectively handle long-term dependencies in sequences, especially when faced with complex, nonlinear dynamic changes, such as changes in bite points over time.

[0136] Modeling Process

[0137] Data preparation:

[0138] The input data is in time series format, usually including multiple image frames and corresponding mechanical data. Each frame of the image contains information such as the position, angle and mechanical stress of the bite point.

[0139] The data is first cleaned and standardized through data preprocessing modules (such as noise filtering, light balancing, etc.) to ensure the quality of input data.

[0140] Feature extraction:

[0141] Features such as the position, shape and motion trajectory of the bite point in the image are extracted from the image data through technologies such as convolutional neural network (CNN).

[0142] The mechanical data (such as applied pressure, strain, etc.) are converted into time series form through the preprocessing module and input into the LSTM network together with the image features.

[0143] LSTM network modeling:

[0144] The time series data (including image features and mechanical data) is input into the LSTM model. The LSTM network continuously updates the memory unit in multiple time steps and outputs the predicted value of the current bite point state (such as position, mechanical stress, etc.).

[0145] The modeling effect can be further enhanced by using a bidirectional LSTM, which enables the network to consider both past and future temporal dependencies.

[0146] Forecasting and Modeling:

[0147] The output of the LSTM network will be used to predict the state change of the bite point in future time steps. This is especially important for modeling dynamic changes, because the bite point may change continuously over time (such as changes in bite force or adjustments in oral structure).

[0148] Through the output of LSTM, we can predict the dynamic evolution of bite points, including predicting their precise locations and changing trends at a certain moment in the future.

[0149] Dynamic change analysis:

[0150] The prediction results of the LSTM model can be used to model and analyze the dynamic changes of the bite point. For example, the change amplitude and change trend of the bite point between different time steps can be analyzed, as well as whether there are certain periodic changes (such as changes in the bite rhythm).

[0151] Subsequent processing:

[0152] For complex dynamic data, the temporal convolutional network (TCN) or other time series analysis methods can be combined to further optimize the model performance, especially when dealing with high-frequency changes and complex nonlinear changes.

[0153] The optimization algorithm trains and adjusts the model to improve detection accuracy and generalization ability.

[0154] By passing the detection results to subsequent modules, this module achieves a seamless connection from data analysis to occlusion status detection.

[0155] 1. Model design and architecture

[0156] In this embodiment, the AI-based occlusion detection module adopts the YOLOv8 (You Only Look Once Version 8) framework, combined with a variety of optimization technologies, and the specific design is as follows:

[0157] Model basic framework:

[0158] YOLOv8 is a single-stage target detection framework that uses a deep convolutional neural network for feature extraction, detection, and classification. This embodiment selects YOLOv8 as the basic framework, mainly considering its performance advantage in small target detection.

[0159] Feature Pyramid Network (FPN):

[0160] The feature pyramid network (FPN) is integrated in the feature extraction part of YOLOv8 to extract multi-scale features such as bite points and contact points in dynamic images, ensuring that the detection target can be identified at different resolutions.

[0161] Attention Mechanism Embedding:

[0162] Embed CBAM (Convolutional Block Attention Module) in the backbone of YOLOv8 to enhance the ability to focus on small targets (such as contact points) and key features. CBAM guides the model to prioritize learning important feature areas through channel attention and spatial attention modules.

[0163] 2. Feature extraction and bite point detection

[0164] In this embodiment, the feature extraction and detection process of the occlusion detection model is as follows:

[0165] Feature extraction:

[0166] The preprocessed dynamic image data is input, and the model extracts local and global features in the image layer by layer through a deep convolutional neural network, and combines FPN to extract multi-scale features.

[0167] Target positioning and classification:

[0168] The YOLOv8 detection head is used to locate the bite points and contact points of the extracted features, and the target bounding box and category confidence are output. The classification task divides the targets into three categories: bite points, non-bite points, and abnormal contact points.

[0169] Loss function optimization:

[0170] Use focal loss to handle the class imbalance problem and improve the detection performance of rare categories (such as abnormal contact points);

[0171] Use IoU loss to optimize the target bounding box positioning and reduce the error.

[0172] 3. Dynamic Modeling

[0173] In this embodiment, in order to capture the time series characteristics of the occlusal process, the following dynamic modeling technology is used:

[0174] Bidirectional LSTM (Bi-LSTM):

[0175] The image frame sequence is temporally modeled, and the causal relationship of the dynamic changes of the occlusal point is captured through bidirectional information flow to generate a description of the occlusal features in the time dimension.

[0176] Temporal CNN:

[0177] A one-dimensional temporal convolutional network is used to model the bite force distribution data, extract the dynamic change law of mechanical characteristics, and combine the image features to realize multimodal feature fusion.

[0178] 4. Model training and optimization

[0179] In this embodiment, in order to improve the model detection performance, the following training and optimization strategies are adopted:

[0180] Transfer Learning:

[0181] The YOLOv8 model is pre-trained using general object detection datasets such as COCO, and then the model is fine-tuned using oral image datasets to adapt to the bite point detection task in oral scenes.

[0182] Hyperparameter Optimization:

[0183] Use the Bayesian optimization algorithm to automatically adjust the model's hyperparameters, including learning rate, batch size (BatchSize) and regularization strength, to improve the model's training effect.

[0184] Data Augmentation:

[0185] The training data is expanded using data enhancement techniques such as rotation, cropping, and mirror flipping. At the same time, a generative adversarial network (GAN) is used to generate diverse simulated image data to improve the generalization ability of the model.

[0186] 5. Output results

[0187] In this embodiment, the output results of the AI-based occlusion detection module include:

[0188] Bite point distribution map: displays the detected bite point locations and classification results;

[0189] Contact point bounding box: marks the location information of the contact point in the form of coordinates;

[0190] Force distribution dynamic diagram: Combine dynamic force distribution characteristics to generate time series analysis results;

[0191] Abnormal contact point marking: Output possible abnormal occlusal features.

[0192] All output results are delivered to the real-time dynamic feedback module in a standardized format.

[0193] In this embodiment, the AI-based occlusion detection module receives the normalized data output by the data preprocessing and enhancement module, and after completing the feature extraction and detection tasks, transmits the analysis results to the real-time dynamic feedback and abnormality analysis module. The time alignment of the data is achieved through the timestamp information to ensure the collaborative work between multi-source data.

[0194] Real-time dynamic feedback and abnormal analysis module

[0195] The main function of the real-time dynamic feedback and abnormality analysis module is to receive the occlusal point distribution and force distribution data output by the AI-based occlusal detection module, and provide doctors with immediate occlusal diagnosis support through data visualization, real-time feedback, and abnormal feature analysis. This module achieves dynamic tracking of occlusal status and accurate identification of abnormal patterns through efficient data processing and intuitive visualization, providing an important basis for subsequent corrective treatment or further diagnosis.

[0196] In this embodiment, the functions of the real-time dynamic feedback and abnormality analysis module include:

[0197] Generate occlusal point distribution map and force distribution heat map in real time to intuitively display the dynamic changes during occlusion;

[0198] Abnormal pattern recognition of occlusal data, including detection of eccentric bite, overload and contact anomalies;

[0199] Provides an interactive 3D visualization interface to dynamically replay the patient's bite process and compare historical data.

[0200] Through real-time feedback and abnormal analysis, this module realizes the conversion of test data into diagnostic results, ensuring the efficiency and practicality of the system in clinical use.

[0201] 1. Real-time feedback generation

[0202] In this embodiment, the real-time dynamic feedback module generates the following feedback content by receiving the output data of the occlusion detection module:

[0203] Bite point distribution diagram:

[0204] Based on the detected bite point coordinates and category information, a bite point distribution map is generated to intuitively display the location and type of each bite point (such as normal bite point, abnormal contact point, etc.). Different colors or symbols are used to mark each category of bite point, so that doctors can quickly identify key areas.

[0205] Force distribution heat map:

[0206] The force distribution data collected by the pressure sensor is mapped onto the three-dimensional tooth model to generate a force distribution heat map. The heat map represents the magnitude of the force through a color gradient (such as from blue to red), clearly reflecting the force distribution state and change trend during the occlusion process.

[0207] Dynamic change curve:

[0208] The changes of bite force and contact point over time are plotted as dynamic curves, which helps doctors understand the timing characteristics of the bite process.

[0209] Data processing uses GPU acceleration technology to ensure that feedback content can be calculated and displayed within 50ms, achieving real-time dynamic feedback.

[0210] Abnormal pattern recognition

[0211] In this embodiment, to detect abnormal patterns during occlusion, the module uses a combination of rule analysis and model analysis:

[0212] Eccentric bite detection:

[0213] By analyzing the spatial distribution of bite points, the number of bite points and force distribution on the left and right sides are calculated to see if they are balanced. If the proportion of bite points or force distribution on one side exceeds the preset threshold, it is marked as eccentric bite.

[0214] Overload detection:

[0215] According to the pressure value of each contact point in the force distribution heat map, the area with abnormal force distribution is identified. If the pressure value of a contact point is significantly higher than the surrounding area (exceeding the set ratio, such as more than 2 times), it is marked as an overload area.

[0216] Contact anomaly detection:

[0217] Combined with the classification results output by the AI ​​detection module, abnormal contact points (such as contact points that should not exist or points that do not meet occlusal specifications) are identified and the doctor is prompted to pay attention through visual markings.

[0218] Time Series Analysis:

[0219] Use dynamic change curves to detect the occurrence time and characteristics of abnormal force fluctuations during occlusion, such as drastic changes in force values ​​in a short period of time or abnormal increases or decreases in the number of contact points.

[0220] 3. Data Visualization

[0221] In this embodiment, the real-time dynamic feedback module provides an interactive three-dimensional visualization interface to display the real-time changes of the patient's occlusion and compare with historical data:

[0222] 3D model display:

[0223] Based on the three-dimensional tooth model generated by laser scanning equipment, the bite point distribution and force distribution heat map are superimposed on the three-dimensional model, supporting the model's rotation, scaling and section viewing functions.

[0224] Dynamic playback:

[0225] By loading the image frame sequence, combining the force distribution data and timestamp information, the patient's bite process can be dynamically replayed to show the formation and disappearance of the bite point and the dynamic changes of force distribution.

[0226] Historical data comparison:

[0227] Compare the current occlusal data with the patient's historical data to generate a comparison report to help doctors evaluate treatment effectiveness or disease progression.

[0228] Data Output

[0229] In this embodiment, the output content of the real-time dynamic feedback module includes:

[0230] Bite point distribution map and force distribution heat map;

[0231] Abnormal pattern analysis results, including markers for eccentric bite, overload and contact abnormalities;

[0232] Dynamic change curve chart;

[0233] 3D model overlay data for visual display.

[0234] All output data are passed to the multimodal data fusion module in a standard format to support further comprehensive diagnosis.

[0235] In this embodiment, the real-time dynamic feedback and abnormality analysis module receives the output data of the AI-based occlusion detection module through an interface and generates feedback content. After the feedback results are combined with the data of the multimodal data fusion module, more complex comprehensive diagnosis tasks can be supported. In addition, the visualization results of this module can be directly presented to the doctor to achieve real-time interaction and diagnosis support.

[0236] Multimodal data fusion module

[0237] The main task of the multimodal data fusion module is to integrate and deeply analyze various heterogeneous data such as dynamic image data, bite force distribution data, and patient medical record data. By introducing the coordinate channel attention mechanism and multimodal joint modeling technology, the spatial and temporal alignment and feature interaction of the data are achieved to generate comprehensive diagnostic results. This module is an important link in the system's derivation from single data analysis to overall diagnosis, and is closely connected with the previous module and subsequent diagnostic suggestions.

[0238] In this embodiment, the functions of the multimodal data fusion module include:

[0239] Data alignment and normalization: spatial and temporal alignment of dynamic image data, three-dimensional structure data, and force distribution data;

[0240] Multimodal joint modeling: Combining the Transformer architecture with multi-source feature interaction modeling methods to extract cross-modal correlations;

[0241] Comprehensive diagnosis generation: Analyze the patient's occlusal characteristics through fused multimodal data and generate comprehensive diagnostic suggestions.

[0242] This module provides doctors with global diagnostic support and improves the intelligence level of the system through deep integration of multimodal data.

[0243] 1. Data alignment and normalization

[0244] In this embodiment, in order to realize the integration of multi-source data, the input dynamic image, bite force distribution and medical record data are first aligned and normalized:

[0245] Time Alignment:

[0246] Using the time stamps provided by the timestamp synchronization unit, the dynamic image frame sequence is aligned with the bite force distribution data.

[0247] The accuracy of time alignment reaches millisecond level, ensuring that the image and force distribution have consistent contact point data at the same time point.

[0248] Spatial Alignment:

[0249] The bite point position in the dynamic image data is mapped to the geometric coordinates of the three-dimensional structure data to generate one-to-one corresponding bite point position information.

[0250] The force distribution data were superimposed on the three-dimensional tooth model through coordinate projection technology to construct the spatial distribution map of the occlusal force.

[0251] Medical record data integration:

[0252] Associate the patient's basic information, medical history and treatment records with data collected in real time.

[0253] The data are formatted using a predefined medical record structured template for subsequent model analysis.

[0254] 2. Multimodal Joint Modeling

[0255] In this embodiment, the multimodal data fusion module deeply fuses multi-source data through feature interaction and joint modeling technology:

[0256] Transformer framework:

[0257] The Transformer architecture is used to jointly model dynamic image features, force distribution features, and medical record features.

[0258] Each modal data first generates its own weighted features through the self-attention mechanism, and then realizes cross-modal feature interaction through the multi-head attention mechanism.

[0259] Coding and weighting process

[0260] To explain more clearly how to code and weight data sources, you can choose one of the following two processing schemes, or a combination of the two:

[0261] Solution 1: Encode and weight each data source separately

[0262] Image data encoding: First, the image data is encoded through a convolutional neural network (CNN) to extract spatial feature information. Each frame of the image is processed into a feature vector that represents the characteristics of each area in the image.

[0263] Mechanical data encoding: At the same time, the mechanical data (such as force, pressure, strain, etc.) are encoded using a fully connected network or convolutional layer and converted into corresponding feature vectors to represent the mechanical characteristics at different time steps.

[0264] Coordinate channel attention mechanism: For each modality’s feature vector, the coordinate channel attention mechanism is applied to calculate the respective weight coefficients. This coefficient will weight different channels based on the spatial position and importance of the feature.

[0265] Weighted processing: Weight the encoded feature vector to highlight important features and suppress irrelevant information.

[0266] Solution 2: Data fusion first, then weighted processing

[0267] Feature fusion: First, the image data and mechanical data are fused into a joint feature vector. The two can be combined by concatenation or additive fusion to form a unified feature representation.

[0268] Coordinate channel attention mechanism weighting: Apply the coordinate channel attention mechanism to the fused feature vector, and assign weights to each feature through the attention mechanism. This mechanism will pay attention to the importance of spatial features and channel features in the data and dynamically adjust the weights.

[0269] Weighted feature processing: The weighted features are passed to the subsequent network for decision making and prediction.

[0270] Coordinate channel attention mechanism:

[0271] Embed the coordinate channel attention module in the dynamic image feature extraction to guide the model to accurately model the spatial position characteristics of the bite point;

[0272] Combine contact point information in force distribution data to enhance the correlation of cross-modal features;

[0273] The coordinate channel attention mechanism is designed to handle the fusion of data from different modalities (image data, mechanical data, etc.) to ensure that different types of data can be reasonably weighted to optimize the performance of the model.

[0274] Coordinate channels refer to the relationship between the spatial position (coordinates) and feature dimensions (channels) of data. This mechanism further improves the network's attention to different modalities by focusing on the specific position of each data source and its corresponding feature channel.

[0275] This mechanism can automatically calculate and selectively weight the features of different data sources during the feature learning process, helping the model to highlight the most informative parts of different modal data and avoid interference from irrelevant features.

[0276] Feature fusion strategy:

[0277] Dynamic images and force distribution features are based on the time dimension, and the temporal convolutional network (Temporal CNN) is used to achieve time series fusion of multimodal features.

[0278] The medical record data are embedded through a fully connected network and combined with other modal features to form a multimodal feature vector of the patient's occlusal characteristics.

[0279] 3. Comprehensive diagnosis generation

[0280] In this embodiment, the fused multimodal features are used to generate a comprehensive diagnosis result, which specifically includes:

[0281] Overall score of occlusal abnormality:

[0282] Based on the joint analysis of dynamic imaging, force distribution and medical record data, a comprehensive score reflecting the patient's occlusal health status is generated. The score ranges from 0 to 100, and the lower the score, the higher the risk of occlusal abnormalities.

[0283] Abnormal pattern classification:

[0284] Based on the feature vector output by the model, determine whether the patient has occlusal problems such as eccentric bite, overload or contact abnormality;

[0285] Generate specific abnormal classification labels and attach corresponding diagnostic reasons.

[0286] Diagnostic suggestion generation:

[0287] Combined with previous records of medical history data, preliminary diagnostic recommendations are generated, such as recommendations for bite adjustment, treatment correction, or further imaging examinations.

[0288] 4. Data Output

[0289] In this embodiment, the output content of the multimodal data fusion module includes:

[0290] Comprehensive diagnostic report, including occlusal abnormality score, classification label and diagnostic suggestions;

[0291] 3D occlusal model with dynamic imaging data and force distribution mapping;

[0292] Cross-modal feature vectors for subsequent in-depth analysis or storage.

[0293] In this embodiment, the multimodal data fusion module receives the analysis data and medical record data from the real-time dynamic feedback module, and generates a comprehensive diagnosis result through feature fusion and joint modeling. The output data is passed to the subsequent diagnosis suggestion module to provide support for the doctor's decision-making. At the same time, the module is connected to the system's data storage unit to ensure that all processing results can be recorded and archived.

[0294] Lightweight edge computing and system deployment module

[0295] For the lightweight edge computing and system deployment module, its main task is to ensure that the system can run stably on real-time, efficient and resource-limited embedded devices through model optimization, edge computing deployment and edge-cloud collaboration mechanism. This module realizes efficient collaborative computing between embedded devices and the cloud through model compression, hardware acceleration and task allocation strategy optimization, provides technical support for the system's real-time occlusion detection, and ensures multi-platform adaptability.

[0296] In this embodiment, the functions of the lightweight edge computing and system deployment module include:

[0297] Model compression and optimization: Reduce model computational complexity and storage requirements through pruning, quantization, and knowledge distillation techniques;

[0298] Edge computing deployment: Real-time inference of deep learning models on embedded devices;

[0299] Edge-cloud collaboration mechanism: divides lightweight tasks and heavy computing tasks to achieve efficient collaboration between edge devices and the cloud;

[0300] System multi-platform adaptation: supports flexible deployment and performance optimization in different hardware environments.

[0301] Through modular design, the system can take into account both real-time and computing performance, providing efficient and convenient technical support for clinical operations.

[0302] 1. Model compression and optimization

[0303] In this embodiment, in order to run the deep learning model on a resource-constrained embedded device, the following optimization method is used:

[0304] Model pruning:

[0305] Prune the convolutional layer weights of the deep learning model based on the importance score to remove the neuron connections that contribute less to the inference results;

[0306] The pruning ratio is adjusted through layer-by-layer optimization, which greatly reduces the amount of computation and memory usage while maintaining the model accuracy.

[0307] Model Quantization:

[0308] Use INT8 quantization technology to compress model weights and activation values ​​from 32-bit floating point numbers (FP32) to 8-bit integers (INT8), significantly reducing model storage requirements and computational complexity;

[0309] A symmetric quantization strategy is introduced to restore the model accuracy through post-quantization training (PTQ).

[0310] Knowledge Distillation:

[0311] The high-precision teacher model is used to guide the training process of the lightweight student model, and the knowledge structure of the teacher model is preserved through the distillation loss function.

[0312] Combined with transfer learning technology, the convergence speed of student models on small-scale data sets can be accelerated.

[0313] 2. Edge computing deployment

[0314] In this embodiment, the optimized lightweight model is deployed on an embedded device (such as a portable oral intraocular camera, a mobile tablet device), and the specific implementation includes:

[0315] Hardware Adaptation:

[0316] Embedded devices use embedded GPUs (such as the NVIDIA Jetson series) or NPUs (such as ARM Mali or Google Edge TPU) to accelerate the model’s reasoning process;

[0317] Model runtime environments designed for low-power devices, such as TensorRT or ONNX Runtime.

[0318] Real-time inference optimization:

[0319] Use batch inference mechanism to process multiple frames of data in parallel to improve throughput;

[0320] Introduce a dynamic memory allocation strategy to reduce resource waste and improve reasoning efficiency.

[0321] Local Cache and Response:

[0322] The key inference results of the model are stored in the local cache, allowing doctors to quickly view the intermediate results of occlusal detection;

[0323] Data is exchanged with other hardware devices (such as hospital terminals) through low-latency interfaces (such as USB or Bluetooth).

[0324] 3. Edge-cloud collaboration mechanism

[0325] In this embodiment, in order to achieve efficient allocation of computing tasks and resource utilization, an edge-cloud collaborative computing mechanism is designed:

[0326] Task division strategy:

[0327] Assign detection tasks with high real-time requirements (such as bite point positioning and force distribution analysis) to edge devices for execution;

[0328] Assign tasks with higher computational complexity (such as dynamic modeling and multimodal fusion) to the cloud for processing.

[0329] Edge-cloud data interaction:

[0330] Enable efficient data transmission between edge devices and the cloud through 5G networks or low-latency communication protocols (such as MQTT);

[0331] Data interaction adopts an asynchronous communication mode. While receiving the processing results from the cloud, the edge device continues to perform subsequent tasks, reducing delays.

[0332] Cloud task processing:

[0333] Cloud servers perform heavy computing tasks based on high-performance GPUs or TPUs;

[0334] Provide data storage and backup services to preserve patients' bite test records and diagnosis results for a long time.

[0335] 4. System multi-platform adaptation

[0336] In this embodiment, in order to realize efficient operation of the system on multiple hardware platforms, the following adaptation scheme is designed:

[0337] Cross-platform model format conversion:

[0338] Export models from development frameworks (such as PyTorch or TensorFlow) using the ONNX format and adapt them to different inference engines (such as TensorRT, CoreML, TFLite);

[0339] According to the characteristics of different hardware platforms, the quantization parameters and optimization strategies are adjusted respectively.

[0340] Dynamic configuration on the device side:

[0341] The system dynamically adjusts the model's operating parameters (such as resolution, batch size, etc.) based on the computing power and memory size of the target hardware device;

[0342] Supports fast switching and unified deployment between devices through configuration files.

[0343] 5. Data output and interaction

[0344] In this embodiment, the output of the lightweight edge computing and system deployment module includes:

[0345] Real-time inference results from edge devices (e.g., bite point distribution, force distribution heat map);

[0346] Comprehensive diagnostic reports and multimodal analysis results generated in the cloud;

[0347] Data interaction records, including task allocation status and processing delay information.

[0348] Through standardized interfaces, module output data is directly transmitted to the real-time feedback module and the multimodal fusion module, ensuring closed-loop collaboration of system functions.

[0349] In this embodiment, the lightweight edge computing and system deployment module receives data from the preprocessing module and the AI ​​detection module, and transmits the results to the real-time feedback module after completing the reasoning task. At the same time, the collaborative interaction with the cloud module ensures the processing capability of complex tasks and forms an efficient integration of edge and cloud computing resources.

[0350] Working principle: First, the core data such as dynamic images, three-dimensional structure and bite force distribution are acquired through the multi-source intraoral image acquisition module, and the time series synchronization of multi-source data is achieved through timestamps; the acquired data undergoes noise reduction, illumination balance, geometric correction and data enhancement in the preprocessing and enhancement module to optimize the data quality. The processed data enters the AI-based bite detection module, which extracts the bite point and contact point features through a deep learning model, and combines dynamic modeling to achieve abnormal detection of the bite process. Subsequently, the real-time dynamic feedback and abnormality analysis module generates visualization content such as bite point distribution map and force distribution heat map based on the detection results, and identifies problems such as eccentric bite and overload through abnormal pattern analysis, providing intuitive diagnostic support for doctors. At the same time, the multimodal data fusion module deeply integrates dynamic images, force distribution data and medical record information to generate a comprehensive diagnostic report and provide diagnosis and treatment recommendations. The system's lightweight edge computing and system deployment module supports real-time operation on embedded devices and cloud processing of complex tasks through model optimization and edge-cloud collaboration mechanisms, ensuring that the system works efficiently and stably in resource-limited environments. The sequentiality of data flow between modules, the interactivity of functions and the edge-cloud collaborative design together constitute the complete technical chain of the present invention in the field of real-time occlusion detection, which can effectively meet the precise diagnosis needs of oral medical care.

[0351] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An AI-enhanced real-time occlusion detection system based on intraoral images, characterized in that: include: Multi-source intraoral image acquisition module, used to obtain dynamic oral image data through a high-definition intraoral camera, and record the three-dimensional structure of the tooth surface in combination with a laser scanning device, while collecting the force distribution of the occlusal contact point through a pressure sensor array; The multi-source intraoral image acquisition module includes a time stamp synchronization unit, which uses a unified time stamp to perform multi-source synchronization on the oral image frame sequence and the bite force distribution data; A data preprocessing and enhancement module, connected to the multi-source intraoral image acquisition module, is used to perform noise filtering, illumination balance and geometric distortion calibration on the acquired image data, and generate multi-angle samples through data enhancement methods such as random rotation, cropping and mirror flipping; An AI-based occlusion detection module, connected to the data preprocessing and enhancement module, is used to perform feature extraction, occlusion point detection, and dynamic change modeling on the processed data through a deep learning model; A real-time dynamic feedback and abnormality analysis module, connected to the AI-based occlusion detection module, is used to generate an occlusion point distribution map, a force distribution heat map, and analyze the occlusion data; A multimodal data fusion module, connected to the AI-based occlusion detection module and the real-time dynamic feedback and abnormality analysis module, respectively, for fusing image data, occlusal force data and patient medical record information; The lightweight edge computing and system deployment module is connected to the above modules to optimize the computing efficiency of AI models and support collaborative computing between edge devices and the cloud.

2. The AI-enhanced real-time occlusion detection system based on intraoral images according to claim 1, characterized in that: The multi-source intraoral image acquisition module comprises: High-definition oral endoscope, used to collect dynamic image data with a resolution greater than 2K and a frame rate ≥ 60fps; Laser scanning equipment to capture the three-dimensional structure and surface texture of teeth; An array of micro pressure sensors to collect force distribution at the occlusal contact points; The timestamp synchronization unit is used to add unified time stamps to dynamic images and force distribution data.

3. The AI-enhanced real-time occlusion detection system based on intraoral images according to claim 1, characterized in that: The data preprocessing and enhancement module includes: A noise reduction processing unit connected to the multi-source intraoral image acquisition module, used for performing noise reduction on the image data based on non-local mean filtering technology and spectral domain de-reflection technology; An optical correction unit for adjusting the brightness and contrast of the image based on an adaptive illumination equalization algorithm; A geometric calibration unit, used to correct the geometric distortion of the image through a feature point matching algorithm; The data enhancement unit is used to randomly rotate, crop and mirror-flip the rectified image data to generate diversified image samples.

4. The AI-enhanced real-time occlusion detection system based on intraoral images according to claim 1, characterized in that: The AI-based occlusion detection module includes: A deep learning model, connected to the data preprocessing and enhancement module, for extracting multi-scale features through a feature pyramid network; Attention mechanism module, embedding convolutional block attention module to enhance the detection ability of small objects; Dynamic modeling unit, used for temporal modeling of image frame sequences based on bidirectional LSTM and temporal convolutional networks; The classification and localization unit is used to optimize the category distribution of bite point detection based on focal loss and optimize the localization of the object bounding box through IoU loss.

5. The AI-enhanced real-time occlusion detection system based on intraoral images according to claim 1, characterized in that: The AI-based occlusion detection module also includes: Transfer learning module, which is used to pre-train on a general object detection dataset and adapt to oral imaging data through fine-tuning; Adaptive hyperparameter optimization module, which is used to adjust the learning rate, batch size, and regularization parameters through the Bayesian optimization algorithm.

6. The AI-enhanced real-time occlusion detection system based on intraoral images according to claim 1, characterized in that: The real-time dynamic feedback and abnormality analysis module includes: A real-time feedback unit connected to the AI-based occlusion detection module for generating an occlusion point distribution map and a force distribution heat map; An abnormality analysis unit, used to extract occlusal abnormality features from the test results and generate a diagnosis report; Visual interface unit, used to dynamically display three-dimensional occlusal data and compare and analyze historical data.

7. The AI-enhanced real-time occlusion detection system based on intraoral images according to claim 1, characterized in that: The multimodal data fusion module includes: A data alignment unit, connected to the AI-based occlusion detection module, for uniformly encoding image data and mechanical data through a coordinate channel attention mechanism; Joint modeling unit, used to interactively model imaging data, mechanical characteristics, and medical record data based on the Transformer framework; Fusion analysis unit for generating comprehensive occlusal diagnosis results and correction recommendations.

8. The AI-enhanced real-time occlusion detection system based on intraoral images according to claim 1, characterized in that: The lightweight edge computing and system deployment module includes: A model compression unit, connected to the AI-based occlusion detection module, for compressing the model through pruning and quantization techniques; Knowledge distillation module, used to optimize the lightweight student model through the teacher model; The edge-cloud collaboration unit is used to complete real-time reasoning on edge devices and migrate complex computing tasks to the cloud.

9. The AI-enhanced real-time occlusion detection system based on intraoral images according to claim 1, characterized in that: The detection function of the system completes the real-time detection of dynamic occlusion points and contact points by connecting a multi-source intraoral image acquisition module, a data preprocessing and enhancement module, and an AI-based occlusion detection module.

10. The AI-enhanced real-time occlusion detection system based on intraoral images according to claim 1, characterized in that: The system is connected to the hospital information system through an API interface to achieve sharing and synchronization of occlusal detection data and support cross-device diagnostic record updates.

Images