Dental maxillofacial growth and development evaluation method and system based on image intelligent analysis

Abstract

The invention discloses a dental and maxillofacial growth and development evaluation method and system based on intelligent image analysis, and relates to the technical field of dental and maxillofacial analysis. According to the dental maxillofacial growth and development evaluation method based on image intelligent analysis, a patient head side position color image is adopted as a data acquisition mode, professional X-ray imaging equipment and a standardized head positioning device are not needed, and the hardware investment cost of a medical institution is greatly reduced. Meanwhile, the deep learning modeling model is used for automatic key point labeling and measurement index calculation, dependence on professionals is reduced, personal errors are avoided, and automatic analysis is achieved. Besides, the method supports home use, high-frequency monitoring can be conveniently carried out, the requirement for continuous evaluation in the orthodontic treatment process is met, the whole evaluation process is simple, convenient and rapid, the period is short, the cost is low, efficient batch analysis can be achieved, and the efficiency and accuracy of dental maxillofacial growth and development evaluation are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of dentofacial analysis, and in particular to a method and system for evaluating the growth and development of the dentofacial based on image intelligent analysis. Background Art

[0002] In the field of orthodontics, the accurate evaluation of dentofacial structures plays a crucial role in formulating treatment plans and monitoring treatment effects. The traditional cephalometric technique, as the gold standard for evaluating maxillofacial structures, has long relied on lateral cephalometric X-ray films of patients for systematic measurement and analysis. Through the coordinated action of professional X-ray imaging equipment, standardized head positioning devices, and professional cephalometric analysis software, this technique provides doctors with objective data on craniofacial skeletal development, tooth position, and soft tissue contours.

[0003] However, this traditional method has the following deficiencies and urgently needs to be improved:

[0004] First, it has a strong dependence on equipment, requiring expensive professional equipment and software, which increases the hardware investment and software licensing fees of medical institutions;

[0005] Second, it has high professional technical requirements. The X-ray shooting, anatomical landmark marking, and measurement data interpretation all need to be operated by professional personnel, with a risk of human error and no automation achievable.

[0006] Third, the service accessibility is low. Patients must go to the hospital for examinations, which are restricted by working hours and cannot be monitored frequently, making it difficult to meet the needs of continuous evaluation in orthodontic treatment.

[0007] Fourth, X-ray examinations involve ionizing radiation and are not suitable for frequent use by special populations such as children and pregnant women. Long-term multiple examinations may also increase health risks.

[0008] Fifth, the entire examination process is cumbersome, with a long cycle, high cost, and no efficient batch analysis achievable, severely limiting the popularization and application of cephalometric techniques. In view of the deficiencies of the prior art, the present invention provides a method and system for evaluating the growth and development of the dentofacial based on image intelligent analysis to solve the above problems. Summary of the Invention

[0009] In view of the deficiencies of the prior art, the present invention provides a method and system for evaluating the growth and development of the dental and maxillofacial region based on image intelligent analysis. By using the lateral color image of the patient's head as the data acquisition method, it does not require professional X-ray imaging equipment and standardized head positioning devices, greatly reducing the hardware investment cost of medical institutions. At the same time, an automatic key point annotation and measurement index calculation are carried out by using a deep learning modeling model, reducing the dependence on professionals, avoiding human errors, and realizing automatic analysis. In addition, this method supports home use. Patients do not need to go to the hospital for examination, are not restricted by the working hours of medical institutions, and can conveniently conduct high-frequency monitoring, meeting the need for continuous evaluation during orthodontic treatment. Due to the use of color image technology, ionizing radiation is avoided, which is suitable for special populations such as children and pregnant women, reducing health risks. The entire evaluation process is simple, fast, has a short cycle, low cost, and can achieve efficient batch analysis, significantly improving the efficiency and accuracy of the evaluation of the growth and development of the dental and maxillofacial region, and has important clinical significance and application value.

[0010] To achieve the above objectives, the present invention is realized through the following technical solutions: A method for evaluating the growth and development of the dental and maxillofacial region based on image intelligent analysis, the evaluation method includes the following steps:

[0011] Step S1, data acquisition, obtaining the lateral color image of the patient's head. During the acquisition process, the acquisition process is optimized by integrating a real-time pose estimation and dynamic feedback system;

[0012] Step S2, data standardization processing, performing data standardization on the picture information;

[0013] Step S3, data annotation, annotating several groups of key point coordinates on the obtained color image;

[0014] Step S4, model construction, constructing a deep learning modeling model, performing deep training on the model to form an artificial intelligence network model;

[0015] Step S5, model prediction, importing the test picture into the artificial intelligence network model, and the artificial intelligence network model automatically calculates each key point of the test picture to obtain measurement indicators;

[0016] Step S6, report generation, generating a standardized measurement report according to each measurement indicator.

[0017] Preferably, when optimizing the acquisition process, based on the MediaPipe Face Mesh model in the field of computer vision, the 3D key point coordinates of the patient's head are tracked in real time through a lightweight convolutional neural network. The head deflection angle is calculated using the triangular plane formed by the tip of the nose and the bilateral tragus points. When the detected deflection angle exceeds the clinical experience threshold of ±15°, the system issues a directional prompt instruction. After the model is constructed, model training is carried out. Based on frameworks such as YOLO, the mapping relationship between the "lateral color map of the head - key point coordinates" is learned.

[0018] Preferably, during the model training process, a data augmentation strategy is adopted. The data augmentation strategy includes the following augmentation operations:

[0019] Randomly rotate the pictures to simulate the influence of different shooting angles and improve the adaptability of the model to head pose changes;

[0020] Perform scaling changes on the pictures to simulate the influence of different face shapes and enhance the generalization ability of the model to different face shapes;

[0021] Adjust the brightness of the pictures to adapt to different lighting conditions and improve the robustness of the model.

[0022] Preferably, in step S2, the LetterBox algorithm is adopted to maintain the aspect ratio of the original picture and uniformly output the pictures to 640×640 pixels.

[0023] Preferably, in step S3, 56 key point coordinates are marked on the lateral color map of the patient's head, and three rounds of independent markings are carried out. For each key point coordinate, calculate the distance between any two of the three rounds of independent marking coordinates and normalize it to the size of the entire image. Take the maximum value among the three as the error measure of the key point, and arbitrate the different points to finally establish the key point coordinate standard.

[0024] Preferably, when arbitrating different points, a key point coordinate error checking mechanism is adopted.

[0025] Preferably, in the error checking mechanism, the following formula is used for calculation:

[0026]

[0027] Where, E k represents the percentage of the key point coordinate error, W and H respectively represent the width and height of the image, max(d ij ) represents the maximum value of the distance between the same key point coordinates marked in any two of the three rounds of independent markings. If the key point coordinate error E k ≤1.2%, take the average value of the three groups of coordinates as the final coordinate. If the key point coordinate error E k> 1.2%, automatically mark the coordinates of this key point and trigger manual review.

[0028] Preferably, in step S6, when the report is presented, a three-dimensional head model is superimposed with key points displayed, a measurement value heat map is provided, the degree of abnormality is mapped using colors, and a growth trend prediction curve is provided to provide a basis for long-term treatment management.

[0029] Preferably, in step S4, the performance of the evaluation model is evaluated through cross-validation and leave-one-out validation, multi-center clinical validation is carried out, a model performance monitoring system is established, and key indicators are regularly evaluated.

[0030] The second aspect of the present invention provides a dental and maxillofacial growth and development evaluation system based on image intelligent analysis, which applies the described dental and maxillofacial growth and development evaluation method based on image intelligent analysis, and includes:

[0031] A data acquisition module for obtaining a lateral color image of the patient's head;

[0032] A data standardization processing module for standardizing the picture information;

[0033] A data annotation module for annotating several groups of key point coordinates on the obtained color image;

[0034] A model construction module for constructing a deep learning modeling model, performing deep training on the model, and forming an artificial intelligence network model;

[0035] A model prediction module for importing a test picture into the artificial intelligence network model, and the artificial intelligence network model automatically calculates each key point of the test picture to obtain measurement indicators;

[0036] A report generation module for generating a standardized measurement report according to each measurement indicator.

[0037] The present invention discloses a dental and maxillofacial growth and development evaluation method and system based on image intelligent analysis, and the beneficial effects thereof are as follows:

[0038] 1. This dental and maxillofacial growth and development evaluation method based on image intelligent analysis does not require professional X-ray equipment, and only ordinary intelligent devices can complete the measurement, reducing the dependence on professional hardware, greatly reducing the hardware cost, breaking through the limitation of expensive imaging equipment, making dental and maxillofacial analysis more economical and efficient, compatible with a variety of common imaging devices, such as smartphones, tablets, digital cameras, etc., with strong adaptability, can be widely promoted and applied, realizes the full-process automation from image acquisition to measurement and analysis, reduces manual intervention, and improves efficiency.

[0039] 2. The method for evaluating the growth and development of the dentofacial region based on image intelligent analysis can complete dentofacial measurement without professional training, significantly reducing the usage threshold. By adopting intelligent recognition and calculation methods, it reduces human errors, improves the stability and repeatability of measurement results, supports multiple cephalometric analysis methods, can calculate multiple measurement indicators with one shot, provides comprehensive evaluation data, can meet the needs of different patients, satisfy various clinical diagnosis scenarios, enhance the application value, combines deep learning methods, supports adaptive expansion, and can be further extended to more measurement standards in the future.

[0040] 3. The method for evaluating the growth and development of the dentofacial region based on image intelligent analysis supports home self-detection. Users can complete shooting without going to the hospital, improving convenience, breaking through geographical restrictions, realizing remote oral health monitoring, providing a more flexible health management plan for patients, increasing the detection frequency, facilitating long-term tracking records, and assisting in formulating personalized treatment plans. It takes no more than 2 seconds from picture upload to result output, realizing instant measurement and analysis, improving the diagnosis and treatment efficiency, supporting rapid batch data processing, can analyze multiple images simultaneously, and is applicable to large-scale screening and clinical research. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0042] Figure 1 It is a schematic diagram of the overall method steps of the present invention;

[0043] Figure 2 It is a schematic diagram of the overall method flow of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0045] Embodiments of the present application provide a method and system for evaluating the growth and development of the dentomaxillofacial region based on image intelligent analysis, which solve the problems of strong equipment dependence of traditional cephalometric techniques, increasing hardware and software costs; high professional technical requirements, with human errors and inability to be automated; low service accessibility, limited patient access to the hospital and difficult to monitor frequently; X-rays being radioactive and not suitable for frequent use by special populations; cumbersome inspection processes, long cycles, high costs, and inability to efficiently analyze in batches, severely restricting the popularization and application of this technology.

[0046] To better understand the above technical solutions, the above technical solutions will be described in detail below in conjunction with the accompanying drawings of the specification and specific embodiments.

[0047] Embodiments of the present invention disclose a method for evaluating the growth and development of the dentomaxillofacial region based on image intelligent analysis. According to the attached Figure 1 to the attached Figure 2 As shown, the evaluation method includes the following steps:

[0048] Step S1, data collection. Obtain the lateral color image of the patient's head. During the acquisition process, optimize the acquisition process through an integrated real-time pose estimation and dynamic feedback system; let the patient maintain a natural head posture and sit on the examination chair. The operator uses daily terminals such as smartphones and tablets on one side of the patient's head to take pictures in real time to ensure that clear images containing the complete dentomaxillofacial structure position are obtained. Collect multiple lateral color images of the head at different angles and positions to increase the diversity of data.

[0049] Step S2, data standardization processing. Standardize the picture information; use the Python programming language and the OpenCV library to implement the LetterBox algorithm. For each collected color image, first obtain its original width W and height H, and set the preset pixel to 640×640. Calculate the aspect ratio

[0050] If r>1, then adjust the image width to 640, and scale the height proportionally to Then fill the black areas on the top and bottom of the image to make the height reach 640;

[0051] If r<1, then adjust the image height to 640, scale the width proportionally to W = 640r, and then fill the black areas on the left and right sides of the image to make the width reach 640.

[0052] If r = 1, then directly scale the image to 640×640.

[0053] Step S3, Data Annotation: Mark several groups of key point coordinates on the acquired color images; use professional image annotation tools such as LabelImg. This tool supports manually marking key point coordinates on the image and saving the annotation information.

[0054] The annotation process involves three professionally trained oral medicine professionals performing three rounds of independent annotation on each color image. Mark 56 key point coordinates on the lateral head color image of the patient. These key points include key turning points such as the tip of the nose point, subnasale point, supralabial point, infralabial point, and mental point.

[0055] Step S4, Model Construction: Build a deep learning modeling model, conduct in-depth training on the model to form an artificial intelligence network model; the model construction is based on frameworks such as YOLO, and use deep learning frameworks such as PyTorch to build a model based on YOLOv8. YOLOv8 has efficient object detection capabilities and can quickly and accurately locate key points in the image.

[0056] Divide the annotated dataset into a training set, a validation set, and a test set in a ratio of 7:2:1. Use the training set to train the model, adopt a Stochastic Gradient Descent (SGD) optimizer, set the learning rate to 0.01, and the momentum to 0.9. During the training process, learn the mapping relationship between "lateral head color image - key point coordinates" based on frameworks such as YOLO.

[0057] Step S5, Model Prediction: Import the test pictures into the artificial intelligence network model, and the artificial intelligence network model automatically calculates each key point of the test pictures to obtain measurement indicators; such as the distance between teeth, the length of the jawbone, etc.

[0058] Step S6, Report Generation: Generate a standardized measurement report based on each measurement indicator.

[0059] Use 3D modeling software such as Blender to build a standard three-dimensional head model. Map the predicted key point coordinates onto the three-dimensional head model to achieve the superimposed display of key points on the three-dimensional head model.

[0060] Generate a heat map using color mapping according to the degree of abnormality of the measurement indicators. For example, map the normal measurement value range to green, map the values that exceed the normal range but to a lesser extent to yellow, and map the values that exceed the normal range and to a greater extent to red.

[0061] Collect the historical data of the patient, use time series analysis methods such as the ARIMA model to predict the growth trend of the patient's dentofacial region, generate a growth trend prediction curve, and provide a basis for long-term treatment management.

[0062] When optimizing the acquisition process, based on the MediaPipe Face Mesh model in the field of computer vision, the 3D key point coordinates of the patient's head are tracked in real time through a lightweight convolutional neural network. The head deflection angle is calculated using the triangular plane formed by the tip of the nose and the bilateral tragus points. When the detected deflection angle exceeds the clinical experience threshold of ±15°, the system issues a directional prompt instruction. After the model is constructed, model training is carried out. Based on frameworks such as YOLO, the mapping relationship between "lateral color head image - key point coordinates" is learned.

[0063] During the model training process, a data augmentation strategy is adopted. The data augmentation strategy includes the following augmentation operations:

[0064] Randomly rotate the pictures to simulate the influence of different shooting angles and improve the adaptability of the model to head pose changes. In each training iteration, randomly rotate the input image by an angle within the range of -30° to 30°. Use the getRotationMatrix2D and warpAffine functions of OpenCV to implement image rotation and simulate the influence of different shooting angles.

[0065] Perform scaling transformation on the pictures to simulate the influence of different face shapes and enhance the generalization ability of the model to different face shapes. Randomly scale the image, and the scaling factor ranges from 0.8 to 1.2. Achieve scaling by adjusting the width and height of the image to simulate the influence of different face shapes.

[0066] Adjust the brightness of the pictures to adapt to different lighting conditions and improve the robustness of the model. Randomly adjust the brightness of the image, and the brightness adjustment factor ranges from 0.7 to 1.3. Use the convertScaleAbs function of OpenCV to implement brightness adjustment and adapt to different lighting conditions.

[0067] In step S2, the LetterBox algorithm is adopted to maintain the aspect ratio of the original image and uniformly output the picture to 640×640 pixels.

[0068] In step S3, mark the 56 key point coordinates on the lateral color head image of the patient, and perform three rounds of independent marking. For each key point coordinate, calculate the distance between the coordinates of the three rounds of independent marking pairwise and normalize it to the size of the entire image, and take the maximum value among the three as the error metric of the key point. Arbitrate the difference points to finally establish the standard of the key point coordinates.

[0069] When arbitrating the difference points, a key point coordinate error checking mechanism is adopted.

[0070] In the error checking mechanism, the following formula is used for calculation:

[0071]

[0072] Among them, E k represents the percentage of the key point coordinate error, W and H respectively represent the width and height of the image, and max(d ij ) represents the maximum value of the distance between the same key point coordinates in any two rounds of independent annotation among the three rounds of independent annotation. If the key point coordinate error E k ≤1.2%, the average value of the three groups of coordinates is taken as the final coordinate. If the key point coordinate error E k >1.2%, the key point coordinate is automatically marked and manual review is triggered.

[0073] Through a large number of experiments and verifications, it is found that when the percentage of the key point coordinate error is less than or equal to 1.2%, the results of the three rounds of independent annotation have high consistency, and at this time, the annotation results are considered reliable. Therefore, this method sets 1.2% as the threshold of the key point coordinate error. When the error percentage is less than or equal to 1.2%, this method takes the average value of the three groups of coordinates as the final coordinate; when the error percentage is greater than 1.2%, it is considered that there are large differences in the annotation of this key point, and the key point coordinate needs to be automatically marked and manual review is triggered to ensure the accuracy of the annotation.

[0074] In step S6, when the report is presented, three-dimensional head model superposition key points are displayed, a measurement value heat map is provided, the degree of abnormality is mapped by color, and a growth trend prediction curve is provided to provide a basis for long-term treatment management.

[0075] In step S4, the performance of the evaluation model is evaluated through cross-validation and leave-one-out validation, multi-center clinical validation is carried out, a model performance monitoring system is established, and key indicators are evaluated regularly. 5-fold cross-validation is adopted, the training set is divided into 5 parts, 4 of them are used as training data each time, and 1 part is used as validation data. 5-fold cross-validation is carried out, and the average performance index is taken. At the same time, leave-one-out validation is adopted. Each time, one sample is left out as the test set, and the rest of the samples are used as the training set. It is repeated many times to evaluate the performance of the model.

[0076] Cooperate with multiple oral medical institutions to collect patient data from different sources and conduct multi-center clinical validation. Apply the model to the data of different centers to evaluate the generalization ability of the model in different environments.

[0077] Regularly collect new patient data and re-evaluate the model. Set key indicators such as the positioning accuracy of key points and the error of measurement indicators, evaluate these indicators regularly, and timely detect the performance changes of the model.

[0078] As an implementation method, in solving the problem of insufficient dynamic adaptability in the prior art, this embodiment achieves optimization by integrating real-time posture estimation and dynamic feedback system. Based on the MediaPipe FaceMesh model in the field of computer vision, the coordinates of 68 three-dimensional key points of the patient's head are tracked in real time through a lightweight convolutional neural network, and the head deflection angle is calculated using the triangular plane formed by the nose tip point and the bilateral tragus points. When the deflection angle is detected to exceed the clinical experience threshold of ±15°, the system issues a directional prompt instruction (such as "Please turn your head 5° to the left") through speech synthesis technology, and superimposes an augmented reality projection on the device screen to guide the patient to adjust to the standard Frankfurt plane posture with a virtual reference line. In the continuous shooting process after the posture meets the standard, an improved clarity evaluation algorithm is used-based on the spatial gradient variance calculation of the image Laplacian operator, combined with the key point matching analysis of the scale-invariant feature transform (SIFT), and three optimal images with a clarity peak and a spatial displacement error of less than 1.5mm are selected from the continuous shooting sequence of 10 frames per second, thereby ensuring the spatial consistency and information integrity of the collected data.

[0079] More specifically, in the clinical dental and maxillofacial growth and development monitoring scenario, the dynamic posture guidance system is integrated with the treatment chair to achieve full-process adaptation. When the patient (especially children) shakes their head due to nervousness, the wide-angle RGB-D camera array (such as Intel RealSense D455) deployed on both sides of the treatment chair synchronously captures depth information and color images at a frequency of 60Hz. The real-time posture estimation module built on the MediaPipe Face Mesh model calculates the direction angle of the triangular plane normal vector formed by the bilateral tragus points and the nasion point, combined with the millimeter-level spatial coordinates obtained by the depth sensor, to accurately quantify the head deflection error. When the coronal plane deflection exceeds 8° or the sagittal plane tilt exceeds 5° (in line with the AAO orthodontic diagnostic standard), the tactile motor built into the headrest of the treatment chair will generate gradient vibration feedback, and the backrest display will present a three-dimensional dynamic guidance animation-for example, a virtual cartoon character demonstrates a standard posture, and improves children's cooperation through a behavioral imitation mechanism. For patients with special needs (such as children with cerebral palsy), the system can switch to auxiliary mode: the flexible positioning device carried by the robotic arm (driven by Shape MemoryAlloy) fine-tunes the head position with a contact force of less than 0.5N, ensuring multi-angle data collection under non-invasive conditions.

[0080] As an implementation method, this embodiment constructs a multimodal data processing pipeline to address the robustness defects in the standardization process. First, the contrast-limited adaptive histogram equalization (CLAHE) algorithm is used to perform local contrast limitation (limitation factor λ=3.0) in units of 8×8 pixel grids to eliminate shadow artifacts caused by uneven lighting in the consulting room. Subsequently, the pre-trained U-Net segmentation network is used to accurately locate the pixel areas of occluders such as hair and fingers, and generate binary masks by utilizing the jump connection characteristics in its encoder-decoder structure. On this basis, the EdgeConnect repair model based on the generative adversarial network (GAN) is introduced: in the first stage, the Canny edge detector is used to extract the contour features of the incomplete area, and the edge generator is used to reconstruct the complete anatomical structure boundary; in the second stage, the generator of the U-Net architecture is used to complete the texture details, and the discriminator uses spectral normalization technology to improve training stability. To further enhance the generalization ability of the model, the Blender physical rendering engine was used to simulate the composite light source conditions with a color temperature of 4000-6000K and an intensity of 200-1000lux in the clinic environment to generate a lighting distribution dataset covering 99% of the real scenes.

[0081] Specifically, in the scenario of primary medical institutions, the illumination adaptation system realizes real-time processing through embedded hardware. In view of the mixed light source environment common in clinics in remote areas (such as the interweaving of sunlight and LED ceiling lights), the CLAHE algorithm is optimized and deployed on the NVIDIA Jetson Nano edge computing module: the 640×640 image is divided into 256 8×8 sub-regions, and the processing delay is controlled within 12ms through shared memory technology when each block independently calculates the cumulative distribution function. When unilateral occlusion is detected (such as long hair covering the auricle area), the U-Net occlusion detection model combines the prior anatomical knowledge base (containing 2000 ear morphological data) for probabilistic reasoning. If it is determined that the probability of the tragus point being occluded is >85%, the multimodal compensation mechanism is triggered-the 3D ear model in the previous scan data of the same patient is automatically retrieved, and the ICP algorithm is used to align it with the current image to generate a virtual projection that conforms to the current head posture to fill the missing area.

[0082] As an implementation, in terms of improving the annotation efficiency and accuracy, this embodiment constructs a semi - automated framework based on deep active learning. The pre - annotation engine adopts the HRNet - W18 network architecture, captures the fine features of tooth edges through a hierarchical parallel multi - resolution sub - network, and combines knowledge distillation technology to transfer and learn the expression of maxillofacial anatomical structure features from a large teacher network (HRNet - W48). In terms of the active learning strategy, Monte Carlo Dropout uncertainty estimation is implemented, Bayesian inference is performed on the prediction variance of each key point, and high - uncertainty samples with a variance value greater than 0.25 are selected for expert review. In the differential point arbitration stage, a contrastive learning twin network is constructed, mapping the three - round annotation results to a 128 - dimensional feature space, and automatic clustering is achieved by calculating the cosine similarity threshold: for annotation points with a similarity higher than 0.95, the geometric median is directly taken, for controversial points with a similarity lower than 0.8, an artificial intervention mechanism is triggered, and at the same time, the expert correction results are recorded for online fine - tuning of the model.

[0083] Specifically, for the batch annotation requirements of large - scale chain dental institutions, the semi - automated system improves efficiency through a cloud - based collaborative architecture. The pre - annotation model is updated using a federated learning framework: local servers of each branch regularly extract the feature gradients of desensitized data (such as the curvature distribution of tooth edges), encrypt and upload them to the central server for model aggregation, which not only ensures data privacy but also continuously optimizes the annotation accuracy. When dealing with rare cases (such as Treacher Collins syndrome), the system automatically retrieves similar 3D craniofacial models in the global case database, generates reference annotation proposals through non - rigid registration, significantly reducing the time-consuming for expert annotation.

[0084] As an implementation, to address the generalization ability limitation in small - sample scenarios, this embodiment designs a hybrid enhancement network architecture. A convolutional block attention module (CBAM) is embedded after the path aggregation network (PAN) layer of the YOLOv8 model. The global average pooling and multi - layer perceptron (MLP) in the channel attention mechanism are used to dynamically adjust the feature channel weights, and the spatial attention branch uses a 7×7 convolutional kernel to capture the context association in the tooth overlap area. In terms of the loss function, the improved FocalLoss sets the balance factor α = 0.75 and the modulation factor γ = 2.5, focusing on strengthening the gradient backpropagation intensity of rare categories (such as impacted teeth), and at the same time combines the complete intersection over union (CIoU) loss function to optimize the bounding box regression accuracy. The data augmentation system adopts the StyleGAN2 - ADA framework, converts the latent vector z to the w+ space through the mapping network, controls the intensity of the pathological features of the generated samples, generates a synthetic dataset containing 50 rare maxillofacial deformities, and ensures the anatomical rationality of the generated images through a progressive training strategy.

[0085] Specifically, in the scenario of orthodontic specialty hospitals, the small-sample enhancement system is deeply integrated into the diagnosis and treatment path. For patients undergoing invisible orthodontics, intraoral scan data and facial images are synchronously entered into the system during each follow-up visit. The pathological feature enhancement data generated by StyleGAN2-ADA (such as simulating the alveolar bone remodeling morphology at different orthodontic stages) is fused with the ClinCheck treatment plan through a digital twin engine to predict the risk areas of root resorption that may occur 6 months in advance. When abnormal eruption of the second molar is detected, the system calls the genetic feature analysis module: by comparing the sagittal profile curvature of the parental maxillofacial scan data and combining the SNP locus information in the genome-wide association study (GWAS) database, an individualized intervention time window recommendation is generated.

[0086] As an implementation, for the problem of insufficient dynamic adaptability of the long-term prediction model, this embodiment constructs an incremental growth prediction system. The Hoeffding tree regressor in the online learning framework is used to dynamically adjust the growth curve prediction parameters by calculating the sliding window statistics (mean, variance, autocorrelation coefficient) of the measurement data stream, and the memory occupancy is controlled within 32 KB to achieve mobile deployment. The probability prediction module integrates a Bayesian neural network and uses Monte Carlo Dropout for approximate variational inference to output the mean and standard deviation when predicting the jaw length and constructs a 95% confidence interval. In terms of genetic feature fusion, the global feature vector of the parental maxillofacial scan point cloud is extracted through a pre-trained PointNet++ network, and it is concatenated with the patient's current features using the gated linear unit (GLU) mechanism. The influence intensity of genetic factors is adjusted through learnable gating weights, and finally an individualized prior distribution of growth rate is generated. The incremental update mechanism of this system can automatically integrate new measurement data every quarter and prevent catastrophic forgetting through the elastic weight consolidation (EWC) algorithm to ensure that the model continuously adapts to the patient's growth and development trajectory.

[0087] Specifically, in the scenario of home intelligent monitoring, the incremental prediction system realizes continuous tracking through wearable devices. The micro-sensors (size 3×3×1 mm) integrated in the intelligent dental appliance continuously monitor the bite force distribution and the jaw micro-movement frequency, and the data is synchronized to the mobile phone APP via Bluetooth LE. The Bayesian neural network performs incremental updates every 8 hours. When it is detected that the standard deviation of the mandibular movement trajectory exceeds the baseline value by 15% for 3 consecutive days, a video capture instruction is automatically triggered - the parent takes a standardized lateral profile video with the mobile phone according to the APP instructions, and the cloud model combines dynamic video stream analysis (not just static images) to evaluate the growth acceleration period. For children at high risk of genetic maxillofacial deformities (such as parents with a history of severe skeletal class II), the system generates a customized orthosis design plan 12 - 18 months in advance and realizes the delivery of the orthosis within 48 hours through the 3D printing service station network.

[0088] The second aspect of the present invention provides a craniofacial growth and development evaluation system based on image intelligent analysis, which applies the above-mentioned craniofacial growth and development evaluation method based on image intelligent analysis, and includes:

[0089] A data acquisition module, which is used to obtain the lateral cephalometric color image of the patient; the client obtains the lateral cephalometric color image of the patient through daily terminals such as smart phones and tablets, and transmits the collected image data to the server through the network.

[0090] A data standardization processing module, which is used to perform data standardization on the picture information; after receiving the image data, the server calls the data standardization processing algorithm, uses the LetterBox algorithm to standardize the image, and stores the processed image in the database.

[0091] A data annotation module, which is used to annotate several groups of key point coordinates for the obtained color image; the annotator uses the annotation tool on the client to annotate the image in the database, and the annotation information is stored in the database.

[0092] A model construction module, which is used to construct a deep learning modeling model, perform deep training on the model, and form an artificial intelligence network model; the server uses the deep learning framework to construct and train the model, and regularly updates the model parameters. During the training process, a data augmentation strategy is adopted to improve the generalization ability of the model.

[0093] A model prediction module, which is used to import the test picture into the artificial intelligence network model, and the artificial intelligence network model automatically calculates each key point of the test picture to obtain the measurement index; the client uploads the test picture to the server, and the server calls the trained model to predict the picture, obtains the measurement index, and returns the result to the client.

[0094] A report generation module, which is used to generate a standardized measurement report according to each measurement index. The client generates a standardized measurement report according to the measurement index returned by the server, including the display of key points of three-dimensional head model superposition, the heat map of measurement values, and the growth trend prediction curve, etc.

[0095] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising said element.

[0096] The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and what is described in the above embodiments and the specification is only to illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements fall within the scope of the present invention claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating the growth and development of the dentomaxillofacial region based on intelligent image analysis, characterized in that, The evaluation method includes the following steps: Step S1, data acquisition. Obtain the lateral color image of the patient's head. During the acquisition process, optimize the acquisition process through an integrated real-time pose estimation and dynamic feedback system; Step S2, data standardization processing. Standardize the data of the picture information; Step S3, data annotation. Mark several groups of key point coordinates on the obtained color image; mark several key point coordinates on the lateral color picture of the patient's head, and perform three rounds of independent annotation. For each key point coordinate, calculate the distance between the coordinates of the three rounds of independent annotation pairwise, and normalize it to the size of the entire image, and take the maximum value of the three as the error metric of the key point. Arbitrate the difference points, and finally establish the standard of the key point coordinates; when arbitrating the difference points, adopt a key point coordinate error inspection mechanism; In the error inspection mechanism, the following formula is used for calculation: Among them, E k represents the percentage of the key point coordinate error, W and H respectively represent the width and height of the image, and max(d ij ) represents the maximum value of the distances between the same key point coordinates in any two rounds of the three rounds of independent annotations. If the key point coordinate error E k ≤1.2%, the average value of the three groups of coordinates is taken as the final coordinate. If the key point coordinate error E k >1.2%, the key point coordinate is automatically marked to trigger manual review; Step S4, model construction. Construct a deep learning modeling model, and perform deep training on the model to form an artificial intelligence network model; Step S5, model prediction. Import the test picture into the artificial intelligence network model, and the artificial intelligence network model automatically calculates each key point of the test picture to obtain measurement indicators; Step S6, report generation. Generate a standardized measurement report according to each measurement indicator.

2. The method for evaluating the growth and development of the dentomaxillofacial region based on image intelligent analysis according to claim 1, wherein, After the model is constructed, model training is carried out, and the mapping relationship between "lateral color picture of the head - key point coordinates" is learned based on the YOLO framework.

3. The method for evaluating the growth and development of the dentomaxillofacial region based on image intelligent analysis according to claim 2, wherein, During the model training process, a data augmentation strategy is adopted. At the initial stage of training, mosaic augmentation is used to provide a diverse visual background in a single training instance and enhance the model's learning ability. The data augmentation strategy includes the following augmentation operations: Randomly rotate the picture to simulate the influence of different shooting angles; Perform scaling changes on the picture to simulate the influence of different face shapes; Adjust the brightness of the picture to adapt to different lighting conditions.

4. The method for evaluating the growth and development of the dental and maxillofacial region based on image intelligent analysis according to claim 1, wherein, In step S2, the LetterBox algorithm is used to maintain the aspect ratio of the original picture and uniformly output the picture to a preset pixel.

5. The method for evaluating the growth and development of the dentomaxillofacial region based on image intelligent analysis according to claim 1, wherein, In step S1, when optimizing the acquisition process, based on the MediaPipe Face Mesh model in the field of computer vision, several groups of three-dimensional key point coordinates of the patient's head are tracked in real time through a lightweight convolutional neural network, and the head deflection angle is calculated using the triangular plane formed by the tip of the nose and the bilateral tragus points.

6. The method for evaluating the growth and development of the dentomaxillofacial region based on image intelligent analysis according to claim 5, wherein, When it is detected that the head deflection angle exceeds the clinical experience threshold of ±15°, the system issues a directional prompt instruction.

7. The method for evaluating the growth and development of the dental and maxillofacial region based on image intelligent analysis according to claim 1, wherein Mark 56 key point coordinates on the lateral color picture of the patient's head and perform three rounds of independent annotation.

8. A method for evaluating the growth and development of the dental and maxillofacial region based on image intelligent analysis according to claim 1, characterized in that, In step S6, when presenting the report, a three-dimensional head model is superimposed with key points displayed, a measurement value heat map is provided, the degree of abnormality is mapped using colors, and a growth trend prediction curve is provided to provide a basis for long-term treatment management.

9. The method for evaluating the growth and development of the dental and maxillofacial region based on image intelligent analysis according to claim 1, wherein In step S4, evaluate the model performance through cross-validation and leave-one-out validation, conduct multi-center clinical validation, establish a model performance monitoring system, and regularly evaluate key indicators.

10. A dental and maxillofacial growth and development evaluation system based on image intelligent analysis, which applies a dental and maxillofacial growth and development evaluation method based on image intelligent analysis as described in any one of claims 1-9, characterized in that, Including: A data acquisition module for obtaining the lateral color image of the patient's head; A data standardization processing module for standardizing the data of the picture information; A data annotation module for annotating several groups of key point coordinates on the acquired color image; A model construction module for constructing a deep learning model, performing deep training on the model, and forming an artificial intelligence network model; A model prediction module for importing a test picture into the artificial intelligence network model, and the artificial intelligence network model automatically calculates each key point of the test picture to obtain measurement indicators; A report generation module for generating a standardized measurement report based on each measurement indicator.

Images