Tooth state intelligent analysis method and system based on oral photography

Abstract

The invention discloses an intelligent tooth state analysis method based on oral photography. The method comprises the following steps: S1, taking an intraoral picture; s2, constructing a sample library; s3, a reinforcement learning training model is adopted; S4, model deployment is carried out; and S5, automatically identifying tooth surface pathology: applying the model to the new intraoral picture, automatically analyzing the new intraoral picture by the model, identifying and classifying the tooth surface pathology in the picture, and outputting the tooth pathology type in the picture. According to the tooth state intelligent analysis method based on oral photography, a dental single lens reflex and an intraoral lens are adopted, intraoral photos are shot, a standardized shooting process is matched, subtle pathological changes on the tooth surface can be accurately captured to construct a sample library, reinforcement learning is adopted to train a model, and through a continuous interaction optimization strategy, the tooth state intelligent analysis method based on oral photography is realized. The method adapts to different shooting conditions or pathological feature changes, so that the model can automatically and accurately identify different types of tooth surface pathologies.

Description

Technical Field

[0001] This invention belongs to the field of oral health management technology, and in particular relates to an intelligent analysis method and system for tooth status based on oral photography. Background Technology

[0002] Dental health assessment refers to the process of systematically analyzing and judging the overall health of teeth, gums, and oral cavity through professional examination or intelligent technologies (such as AI, image analysis, etc.). Its core goal is to identify potential problems (such as tooth decay, periodontal disease, malocclusion, etc.) and provide a basis for subsequent treatment or prevention. However, traditional dental examinations mainly rely on the dentist's visual observation and probe palpation, which are subject to experience differences. For example, the missed diagnosis rate of early interproximal caries is as high as 30%, and paper medical records are difficult to track dynamic changes. With the development of intelligent technologies, intelligent assessment methods are gradually replacing traditional dental health assessment methods and becoming the mainstream. Intelligent dental health assessment is a technology that combines artificial intelligence and oral medicine. It provides users with a preliminary assessment of dental health through image analysis, data modeling, and professional knowledge.

[0003] Chinese patent application CN118644443A, published on September 13, 2024, discloses a neural network-based tooth condition assessment method. This method employs a YOLOv8 network model for multi-level feature extraction and fusion, utilizing the CBAM layer to capture key features, thereby achieving automatic detection and category determination of dental lesion areas. However, in practice, due to the use of supervised learning (YOLOv8), the sample dataset and labels often need to be fixed. If a new pathological type (such as a newly discovered dental disease) is added, the data needs to be re-labeled and the model retrained. Furthermore, higher model complexity leads to higher hardware requirements. For example, the YOLOv8 network model in the aforementioned application, especially the version embedding the CBAM layer, has high computational complexity and requires strong hardware support (such as high-performance GPUs) for training and inference. It also requires exporting ONNX files for cross-platform deployment, limiting its application in resource-constrained environments. Moreover, the complex model structure also results in a large model size, which is not conducive to deployment on mobile devices or embedded systems. Summary of the Invention

[0004] To overcome the shortcomings of the prior art, this invention provides an intelligent tooth condition analysis method based on oral photography. By using reinforcement learning to train the recognition model and continuously optimizing the strategy through interaction, the model can adapt to different shooting conditions or changes in pathological features, enabling it to automatically and accurately identify different types of tooth surface pathology.

[0005] According to one aspect of the present invention, a method for intelligent analysis of tooth state based on oral photography is provided, comprising: Obtain several intraoral photographs; Based on several intraoral photographs, a labeled sample library was formed; A reinforcement learning framework is constructed, which treats the dental state analysis scenario as a reinforcement learning environment. Each intraoral photograph is used as the state input, the model's recognition result is used as the action output, and the recognition accuracy or the degree of matching of pathological types is used as the reward signal. Based on deep reinforcement learning architectures such as Double DQN, Dueling DQN, or Rainbow DQN, construct a CNN model that includes attention mechanisms (such as Squeeze-and-Excitation Networks, SE Net, or Convolutional Block Attention Module, CBAM), add an SE module or CBAM layer after each convolutional block, and then combine it with a fully connected layer for decision-making; The constructed recognition model is trained using a sample library. During the training process, the recognition model receives input images and outputs pathology type predictions, while adjusting the model parameters based on the recognition accuracy or the degree of matching of pathology types. The trained recognition model is deployed on the target device to perform intelligent analysis of the tooth condition on the input intraoral photos and output the dental case type in the photos.

[0006] As a further technical solution, the constructed recognition model is trained, including: Randomly initialize the weights and biases of the recognition model; Design a multi-level reward function that comprehensively considers recognition accuracy, pathological type severity, and recognition speed; Using Curriculum Learning, the model is first trained to recognize simple pathologies, and then the complexity is gradually increased. By continuously receiving new intraoral photographs and attempting to identify them, rewards or penalties are given based on the accuracy of identification or the degree of matching with the pathological type. Adjust model parameters based on reward signals to optimize the recognition strategy; Repeat the interactive learning and policy optimization process until the model reaches the expected recognition accuracy or reaches the predetermined number of training rounds.

[0007] As a further technical solution, the method also includes: The recognition model is packaged into an executable program or API interface and deployed on a local computer, server or cloud platform.

[0008] As a further technical solution, the method also includes: The identification model continuously receives new intraoral photographs and makes predictions. It adjusts its strategy based on the difference between the prediction results and the actual pathological types. At the same time, it introduces online learning or transfer learning techniques. When encountering new pathological types, it can quickly adapt to new pathological features using existing knowledge without re-labeling data and training the entire model.

[0009] As a further technical solution, several intraoral images are obtained, including: The images were taken from six standard perspectives according to the FDI tooth position zoning method, including frontal images of anterior teeth, left / right occlusal images, upper / lower dental arch images, and close-up images of single teeth, with the error controlled within the preset angle range.

[0010] As a further technical solution, lighting control during the shooting process includes: A ring flash is used to provide uniform, shadowless illumination, and the ring flash is used to provide side lighting at a set angle to eliminate reflections. A polarizing filter is used to eliminate glare in highly reflective areas.

[0011] As a further technical solution, the method also includes: The test determines whether the intraoral photographs taken meet the following requirements: the tooth surface occupies more than a set proportion of the image, the clarity meets the set requirements, and there are no missing key teeth. If any requirement is not met, a retake prompt will be triggered.

[0012] According to one aspect of the present invention, a dental state intelligent analysis system based on oral photography is provided, comprising an imaging device and a processing device. The imaging device is used to acquire a plurality of intraoral photographs and transmit them to the processing device, and the processing device is used to execute the aforementioned dental state intelligent analysis method based on oral photography.

[0013] As a further technical solution, the processing device includes: The first main module is used to form a labeled sample library based on several intraoral photographs; The second main module is used to build a reinforcement learning framework, which treats the tooth state analysis scenario as a reinforcement learning environment. Each intraoral photograph is used as the state input, the model's recognition result is used as the action output, and the recognition accuracy or the degree of matching of pathological types is used as the reward signal. The third main module is used to select reinforcement learning algorithms and build recognition models according to task requirements; The fourth main module is used to train the constructed recognition model using a sample library. During the training process, the recognition model receives input images and outputs pathology type predictions, while adjusting the model parameters according to the recognition accuracy or the degree of matching of pathology types. The fifth main module is used to deploy the trained recognition model on the target device, perform intelligent analysis of the tooth status of the input intraoral photos, and output the dental case type in the photos.

[0014] According to one aspect of the present invention, a non-transitory computer-readable storage medium is provided, the non-transitory computer-readable storage medium storing computer instructions that cause the computer to execute the aforementioned intelligent analysis method for tooth state based on oral photography.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention employs a dental-grade SLR camera and an intraoral lens to capture intraoral photographs. Combined with a standardized shooting process, it accurately captures subtle pathological changes on the tooth surface to build a sample library. Based on this library, reinforcement learning is used to train the model. Through continuous interactive optimization strategies, the model adapts to different shooting conditions or changes in pathological features, enabling it to automatically and accurately identify different types of tooth surface pathologies such as cavities, tooth cracks, tartar buildup, and tooth discoloration. Furthermore, online learning or transfer learning techniques are introduced, allowing the model to quickly adapt to new features when encountering new pathological types, without needing to re-label data and retrain the model, further improving efficiency and facilitating rapid tooth condition assessment results.

[0016] 2. This invention does not rely on high-performance computing resources. The model can be deployed on local devices or in the cloud. By utilizing the elastic computing and storage capabilities of cloud computing platforms, the model can achieve rapid response and efficient processing. Overall, the hardware requirements are low, making it particularly suitable for environments with limited resources. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of a method for intelligent analysis of tooth status based on oral photography, provided in an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram illustrating the construction and training of the identification model in an intelligent tooth state analysis method based on oral photography, provided in an embodiment of the present invention. Detailed Implementation

[0020] It should be noted that existing technologies using supervised learning require re-labeling data and retraining the model when adding new pathology types. This places high demands on hardware, results in large model sizes, and hinders deployment. These are all derivative problems of existing model limitations, stemming from the fact that "higher model complexity leads to higher hardware requirements," and that existing technologies "require exporting ONNX files for cross-platform deployment," coupled with the large model size due to complex model structures. The root cause of these problems lies in the fact that existing technologies often employ highly complex models, such as the YOLOv8 network model in the comparison file, and embed CBAM layers. Therefore, the core design of this invention is to abandon highly complex models and eliminate the need for re-labeling data and retraining the model when adding new pathology types. Furthermore, this invention utilizes the elastic computing and storage capabilities of cloud computing platforms to achieve rapid model response and efficient processing.

[0021] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units, not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0023] like Figures 1-2 As shown in the figure, an intelligent tooth state analysis method based on oral photography provided by an embodiment of the present invention includes the following steps: S1. Take photos inside the mouth: Using a dental DSLR camera with an intraoral lens, intraoral photographs are taken to record the condition of the teeth and capture subtle pathological changes, such as cavities and microcracks, providing basic data for subsequent tooth condition assessment. Before this step, patients brush and floss to avoid food debris interference (e.g., in assessing proximal caries), and a disposable oral mirror retractor (such as the Hu-Friedy OptraGate) is used to expose the entire dentition, avoiding lip / cheek obstruction.

[0024] In this embodiment, the SLR camera was configured as a Canon EOS 5D Mark IV with a 100mm macro lens, and the intraoral lens was configured as a Canon Macro Ring Lite MR-14EX Ⅱ. Parameter settings included: 1) Resolution ≥ 30 megapixels (clear display of 50μm-level details for each tooth), RAW format storage, ISO ≤ 400 to reduce noise; 2) Color temperature control: 5500K standard white light, color rendering index (CRI) ≥ 90, specifically using a gray card to calibrate white balance, such as X-Rite ColorChecker. Six standard viewing angles were taken according to the FDI tooth position zoning method, including frontal images of anterior teeth, left / right occlusal images, upper / lower dental arch images, and close-up images of a single tooth, with errors controlled within ±5°. Referring to Table 1, the viewing angles and their operational points are as follows: Frontal view of anterior teeth: The lens is perpendicular to the incisal edge, with a focal length of 30cm, mainly used to screen for incisal edge wear, enamel cracks, etc. Left / right occlusal image: The lens is at a 45° angle to the occlusal plane and includes the upper and lower dentition. It is mainly used to screen for proximal caries, tartar, etc. Upper / lower dental arch image: The lens faces the palatal / lingual side, with depth of field covering the entire dental arch. It is mainly used to screen for dental crowding, marginal leakage of restorations, etc. Single tooth close-up image: macro mode (1:1 magnification ratio), focal length 10cm, mainly used for screening hidden cracks, wedge-shaped defects, etc. Lighting control during the shooting process includes: using a ring flash to provide uniform, shadowless illumination, and using the ring flash to provide 10° side lighting to eliminate reflections (such as enamel specular reflection interference), and using a polarizing filter to eliminate glare in highly reflective areas (such as metal restorations).

[0025] Table 1 Shooting Angle and Key Operational Points .

[0026] The quality requirements for intraoral photographs in this step are as follows: A. The surface of the teeth occupies more than 60% of the image; B. Detect sharpness through edge sharpness algorithm (e.g., MTF50 value > 50 lp / mm). C. Use AI to automatically check if key teeth are missing. If a tooth (such as a wisdom tooth) is not in the frame, a retake prompt will be triggered.

[0027] S2. Constructing the sample library: The intraoral photographs were organized and labeled with the types of dental pathology in the photographs (such as cavities, tooth cracks, tartar buildup, tooth discoloration, etc.) to form a labeled sample library, which provides training data for reinforcement learning models, enabling the models to learn and identify different types of dental surface pathology.

[0028] In this step, annotation tools (such as LabelImg, VGG Image Annotator, etc.) are used for manual or semi-automatic annotation. Dental pathology types include, but are not limited to: Cavities: Hole-like defects on the tooth surface caused by bacterial action. The demineralized enamel area appears chalky white / brownish-black, ICDAS grade 1-6 (white spots → cavities). Tooth cracks: Fine cracks appear on the tooth surface, presenting as mesial and distal linear dark lines, which may be caused by factors such as occlusal trauma and wear. Tartar buildup: Hard deposits that form on the surface of teeth, mainly composed of calcified dental plaque and food debris; Tooth discoloration: Color changes on the surface of teeth caused by smoking, drinking tea, drinking coffee, etc.

[0029] S3. Employ reinforcement learning to train the model: S31. Define a reinforcement learning framework: Treat the tooth condition assessment scenario as a reinforcement learning environment, where each intraoral photograph is taken as the state input, the model's recognition result is the action output, and the recognition accuracy or the degree of matching of pathological types is the reward signal; In this embodiment, a reinforcement learning agent is designed to automatically execute this step. This agent can receive intraoral photographs as input, process them through an internal model, and output the recognition result of tooth surface pathology.

[0030] S32. Model Selection and Construction: Choosing a deep reinforcement learning architecture: Based on the task requirements, i.e. the complexity of the problem and the characteristics of the data, select a suitable deep reinforcement learning architecture, including but not limited to Double DQN, Dueling DQN or Rainbow DQN; Build a neural network model: Build a CNN model that includes attention mechanisms (such as Squeeze-and-Excitation Networks, SE Net, or Convolutional Block Attention Module, CBAM), add an SE module or CBAM layer after each convolutional block, and then combine it with a fully connected layer for decision-making.

[0031] In this embodiment, Rainbow DQN deep learning architecture is specifically chosen to build the neural network model. The model construction process is as follows: 1) Input layer design: The input layer receives a pre-processed intraoral photograph as input. Assuming the photograph is resized to a fixed size, such as 256x256 pixels, and is an RGB color image, the input layer will be a tensor of size 256x256x3.

[0032] 2) Convolutional Neural Network (CNN) section: Multiple convolutional layers are used to extract image features. Each convolutional layer contains multiple convolutional kernels (filters) that slide across the image and compute dot products to extract different features, such as edges and textures.

[0033] Convolution block 1: Convolutional layer: Contains 32 3x3 convolutional kernels with a stride of 1 and padding of "same" to maintain the feature map size.

[0034] Activation function: The ReLU activation function is used to introduce nonlinearity.

[0035] Pooling layer: 2x2 max pooling with a step size of 2, used for dimensionality reduction.

[0036] Convolutional block 2: Convolutional layer: Contains 64 3x3 convolutional kernels, with a stride of 1 and padding of "same".

[0037] Activation function: ReLU.

[0038] Pooling layer: 2x2 max pooling with a step size of 2.

[0039] Attention mechanism: Add an attention mechanism module, such as Squeeze-and-ExcitationNetworks (SE Net) or Convolutional Block Attention Module (CBAM), after each convolutional block to enhance the model's attention to important features.

[0040] 3) Fully connected layer: After the CNN part, the extracted feature map is flattened into a one-dimensional vector.

[0041] Fully connected layer 1: Contains 128 neurons and uses the ReLU activation function to further process features.

[0042] Fully connected layer 2 (output layer): The number of neurons is equal to the number of possible tooth surface pathology types (N), and no activation function is used or linear activation is used to output the Q value for each action (pathology type).

[0043] 4) Rainbow DQN specific components: Rainbow DQN combines several improved DQN techniques, including Double DQN, Dueling DQN, Prioritized Experience Replay, Multi-step Learning, Distributional DQN, and NoisyNets.

[0044] Double DQN: Reduces overestimation by decoupling the calculation of the target Q value from the action selection.

[0045] Dueling DQN: Decomposes Q-value networks into state value functions and advantage functions to improve learning efficiency.

[0046] Prioritized Experience Replay: Samples are sampled based on their importance (TD error) to improve sample utilization.

[0047] Multi-step Learning: Using multi-step rewards to accelerate the learning process.

[0048] Distributional DQN: Learns the distribution of Q-values ​​rather than their expected values, thus improving stability.

[0049] Noisy Nets: Adaptive exploration is achieved by replacing the traditional ε-greedy strategy with parameterized noise.

[0050] In summary, the model structure constructed in this embodiment is as follows: Input layer: 256x256x3 (RGB image) Convolution block 1: Convolutional layer: 32 3x3 convolutional kernels, stride 1, padding "same". Activation function: ReLU Pooling layer: 2x2 max pooling, with a stride of 2. Convolutional block 2: Convolutional layer: 64 3x3 convolutional kernels, stride 1, padding "same". Activation function: ReLU Pooling layer: 2x2 max pooling, with a stride of 2. Attention mechanism: SE module or CBAM layer Flattening layer: flattens the feature map into a one-dimensional vector. Fully connected layer 1: 128 neurons, ReLU activated Fully connected layer 2 (output layer): N neurons (N is the number of pathological types), with no activation function or linear activation. Rainbow DQN components: Double DQN, Dueling DQN, Prioritized Experience Replay, Multi-step Learning, Distributional DQN, Noisy Nets.

[0051] S33. Model Training: The model is trained using data from the sample database, which is divided into a training set and a test set. The training set is used to train the model, enabling it to learn the characteristics and recognition methods of tooth surface pathology. The test set is used to evaluate the model's performance, ensuring that the model maintains a high recognition accuracy even on unseen data. During training, the model interacts with the environment (i.e., receives input images and outputs pathology type predictions) and adjusts its strategy (i.e., model parameters) based on feedback (such as prediction accuracy) to optimize recognition performance. The specific operations of this sub-step are as follows: Initialize model parameters: Randomly initialize the weights and biases of the neural network; Design a reward mechanism: Design a multi-level reward function that comprehensively considers recognition accuracy, pathological type severity, and recognition speed; Introducing the course: Using Curriculum Learning, the model is first trained to recognize simple pathologies, and then the complexity is gradually increased; Interactive learning: Through continuous interaction with the environment, i.e., constantly receiving new intraoral photos and attempting to recognize them, rewards or penalties are given based on the accuracy of the recognition results; Strategy optimization: Model parameters are adjusted based on the reward signal to optimize the recognition strategy. In DQN, action selection is optimized by updating the Q-value. This embodiment uses temporal difference error (TD Error) to update the Q-value, and the process is as follows: TD Target: ; In the formula, For instant rewards; This is a discount factor used to weigh current and future rewards; For the next state The maximum Q-value is calculated through the target network, with parameters as follows: ; TD error: ; Measure the difference between the current Q value and the target Q value; Q value update: Minimize the mean squared error loss using gradient descent: L ( θ ) = E [ ( Target-Q ( s,a; θ ) ) 2 ] .

[0052] Furthermore, the optimization strategy for action selection is an ε-greedy strategy, specifically: exploring randomly with probability ε, and selecting the current optimal action with 1−ε. The goal is to balance exploration (trying new actions) with exploitation (selecting the known best actions). Initially, a higher ε (e.g., 1.0) is set, which is then gradually decreased (e.g., 0.01) to favor exploitation.

[0053] In this embodiment, the overall process of strategy optimization is as follows: Initialization: Main Network and target network .

[0054] Interactive environment: Choose actions based on the ε-greedy strategy. .

[0055] Perform the action and observe Stored in the playback buffer.

[0056] train: Sample a batch of data from the buffer.

[0057] Calculate the TD target and update the main network parameters. .

[0058] Periodically synchronize target network parameters .

[0059] Repeat until the Q-value converges or the maximum number of steps is reached.

[0060] Iterative training: Repeat the above interactive learning and policy optimization process until the model achieves a satisfactory recognition accuracy or reaches the predetermined number of training rounds.

[0061] This step enables the model to automatically identify and classify tooth surface pathologies, and to adapt to different imaging conditions or changes in pathological features through continuous interactive optimization strategies.

[0062] S4. Model Deployment: The trained model is deployed on local devices or in the cloud to support actual clinical diagnosis. Specifically, the model is packaged into an executable program or API interface for integration with other systems and deployed on the local computer, server, or cloud platform of the dental clinic. By utilizing the elastic computing and storage capabilities of the cloud computing platform, the model can achieve rapid response and efficient processing. Doctors or patients can obtain intelligent assessment results of the tooth condition by uploading intraoral photos to the platform.

[0063] S5. Automatic identification of tooth surface pathology: The deployed model is then applied to new intraoral photographs. The model automatically analyzes these photographs, identifies and classifies surface pathologies of the teeth, and outputs the pathological type. In this step, the model continuously receives new intraoral photographs and makes predictions. It adjusts its strategy based on the difference between the predicted and actual pathological types. Simultaneously, online learning or transfer learning techniques are introduced. When encountering new pathological types, the model quickly adapts to the new pathological features using existing knowledge through online learning or transfer learning, without needing to re-label data and retrain the entire model, further improving the model's adaptability and efficiency.

[0064] Based on the same inventive concept as the aforementioned method embodiments, this embodiment of the invention also provides a tooth state intelligent analysis system based on oral photography, including an imaging device and a processing device. The imaging device is used to acquire several intraoral photographs and transmit them to the processing device, and the processing device is used to execute the aforementioned tooth state intelligent analysis method based on oral photography.

[0065] The processing device in this embodiment is implemented through programmed processing using a device with processor functionality. Therefore, in practical engineering, its functionality is encapsulated into various modules, including: The first main module is used to form a labeled sample library based on several intraoral photographs; The second main module is used to build a reinforcement learning framework, which treats the tooth state analysis scenario as a reinforcement learning environment. Each intraoral photograph is used as the state input, the model's recognition result is used as the action output, and the recognition accuracy or the degree of matching of pathological types is used as the reward signal. The third main module is used to select reinforcement learning algorithms and build recognition models according to task requirements; The fourth main module is used to train the constructed recognition model using a sample library. During the training process, the recognition model receives input images and outputs pathology type predictions, while adjusting the model parameters according to the recognition accuracy or the degree of matching of pathology types. The fifth main module is used to deploy the trained recognition model on the target device, perform intelligent analysis of the tooth status of the input intraoral photos, and output the dental case type in the photos.

[0066] This invention provides an intelligent tooth condition analysis system based on oral photography. It employs a dental-grade SLR camera and an intraoral lens to capture intraoral photographs, constructing a sample library. Reinforcement learning is then used to train the model, which adapts to different shooting conditions or pathological features through continuous interactive optimization strategies. This enables the model to automatically and accurately identify different types of tooth surface pathologies, such as cavities, tooth cracks, tartar buildup, and tooth discoloration. Furthermore, online learning or transfer learning techniques are introduced, allowing the model to quickly adapt to new features when encountering new pathological types without needing to re-label data and retrain the model, further improving efficiency and facilitating rapid tooth condition assessment.

[0067] Based on the same inventive concept as any of the foregoing embodiments, this embodiment of the invention also provides a non-transitory computer-readable storage medium storing computer instructions that cause the computer to execute the aforementioned intelligent analysis method for tooth status based on oral photography.

[0068] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0069] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. 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 illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0070] 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.

[0071] 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.

[0072] In summary, this invention discloses an intelligent tooth condition analysis method based on oral photography, relating to the field of oral health management technology. The method includes the following steps: S1, taking intraoral photographs; S2, constructing a sample library; S3, training the model using reinforcement learning; S4, model deployment; and S5, automatically identifying tooth surface pathologies: applying the model to new intraoral photographs, automatically analyzing the new photographs, identifying and classifying tooth surface pathologies in the photographs, and outputting the type of tooth pathology in the photographs. This intelligent tooth condition analysis method based on oral photography uses a dental-grade SLR camera with an intraoral lens to take intraoral photographs. Combined with a standardized shooting process, it can accurately capture subtle pathological changes on the tooth surface to construct a sample library. Reinforcement learning is used to train the model, and through continuous interactive optimization strategies, it adapts to different shooting conditions or changes in pathological features, enabling the model to automatically and accurately identify different types of tooth surface pathologies.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention 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 or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent analysis of tooth condition based on oral photography, characterized in that, include: Obtain several intraoral photographs; Based on several intraoral photographs, a labeled sample library was formed; A reinforcement learning framework is constructed, which treats the dental state analysis scenario as a reinforcement learning environment. Each intraoral photograph is used as the state input, the model's recognition result is used as the action output, and the recognition accuracy or the degree of matching of pathological types is used as the reward signal. Select a reinforcement learning algorithm based on task requirements and construct a recognition model; Based on deep reinforcement learning architectures such as Double DQN, Dueling DQN, or Rainbow DQN, construct a CNN model that includes an attention mechanism, add an SE module or CBAM layer after each convolutional block, and then combine it with a fully connected layer for decision-making; The constructed recognition model is trained using a sample library. During the training process, the recognition model receives input images and outputs pathology type predictions, while adjusting the model parameters based on the recognition accuracy or the degree of matching of pathology types. The trained recognition model is deployed on the target device to perform intelligent analysis of the tooth condition on the input intraoral photos and output the dental case type in the photos.

2. The intelligent tooth state analysis method based on oral photography according to claim 1, characterized in that, The constructed recognition model is trained, including: Randomly initialize the weights and biases of the recognition model; Design a multi-level reward function that comprehensively considers recognition accuracy, pathological type severity, and recognition speed; Using Curriculum Learning, the model is first trained to recognize simple pathologies, and then the complexity is gradually increased. By continuously receiving new intraoral photographs and attempting to identify them, rewards or penalties are given based on the accuracy of identification or the degree of matching with the pathological type. Adjust model parameters based on reward signals to optimize the recognition strategy; Repeat the interactive learning and policy optimization process until the model reaches the expected recognition accuracy or reaches the predetermined number of training rounds.

3. The intelligent tooth state analysis method based on oral photography according to claim 1, characterized in that, The method further includes: The recognition model is packaged into an executable program or API interface and deployed on a local computer, server or cloud platform.

4. The intelligent tooth state analysis method based on oral photography according to claim 3, characterized in that, The method further includes: The identification model continuously receives new intraoral photographs and makes predictions. It adjusts its strategy based on the difference between the prediction results and the actual pathological types. At the same time, it introduces online learning or transfer learning techniques. When encountering new pathological types, it can quickly adapt to new pathological features using existing knowledge without re-labeling data and training the entire model.

5. The intelligent tooth state analysis method based on oral photography according to claim 1, characterized in that, Obtain several intraoral photographs, including: The images were taken from six standard perspectives according to the FDI tooth position zoning method, including frontal images of anterior teeth, left / right occlusal images, upper / lower dental arch images, and close-up images of single teeth, with the error controlled within the preset angle range.

6. The intelligent tooth state analysis method based on oral photography according to claim 5, characterized in that, Lighting control during filming includes: A ring flash is used to provide uniform, shadowless illumination, and the ring flash is used to provide side lighting at a set angle to eliminate reflections. A polarizing filter is used to eliminate glare in highly reflective areas.

7. The intelligent tooth state analysis method based on oral photography according to claim 5, characterized in that, The method further includes: The test determines whether the intraoral photographs taken meet the following requirements: the tooth surface occupies more than a set proportion of the image, the clarity meets the set requirements, and there are no missing key teeth. If any requirement is not met, a retake prompt will be triggered.

8. A smart tooth condition analysis system based on oral photography, characterized in that, The device includes an imaging device and a processing device. The imaging device is used to acquire several intraoral photographs and transmit them to the processing device. The processing device is used to execute the intelligent tooth state analysis method based on oral photography as described in any one of claims 1 to 7.

9. The intelligent tooth state analysis system based on oral photography according to claim 8, characterized in that, The processing device includes: The first main module is used to form a labeled sample library based on several intraoral photographs; The second main module is used to build a reinforcement learning framework, which treats the tooth state analysis scenario as a reinforcement learning environment. Each intraoral photograph is used as the state input, the model's recognition result is used as the action output, and the recognition accuracy or the degree of matching of pathological types is used as the reward signal. The third main module is used to select reinforcement learning algorithms and build recognition models according to task requirements; The fourth main module is used to train the constructed recognition model using a sample library. During the training process, the recognition model receives input images and outputs pathology type predictions, while adjusting the model parameters according to the recognition accuracy or the degree of matching of pathology types. The fifth main module is used to deploy the trained recognition model on the target device, perform intelligent analysis of the tooth status of the input intraoral photos, and output the dental case type in the photos.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to execute the intelligent tooth state analysis method based on oral photography as described in any one of claims 1 to 7.

Images