Oral disease early screening system based on artificial intelligence auxiliary diagnosis system

Abstract

The invention provides an oral disease early screening system and screening method based on an artificial intelligence auxiliary diagnosis system. According to the system, oral images and medical information of a patient are collected through a screening tool, an AI model is trained for auxiliary diagnosis, and intelligent services are provided for doctors in the diagnosis system. Artificial intelligence is introduced into oral medicine, the oral disease diagnosis efficiency and accuracy can be remarkably improved, support is provided for intelligent medical treatment, remote medical treatment and the like, and good application prospects are achieved. The method is suitable for the cross field of oral medicine and artificial intelligence, can be applied to hospitals, physical examination institutions and the like, and realizes early screening and early diagnosis of oral diseases.

Description

Technical Field

[0001] The present invention relates to the field of medical artificial intelligence, and in particular to an early screening system for oral diseases based on an artificial intelligence-assisted diagnosis system, which belongs to artificial intelligence-assisted diagnosis technology. Background Art

[0002] Oral diseases are common and frequently occurring diseases. If common oral diseases such as caries and periodontitis are not diagnosed and treated in time, they can lead to tooth loss and affect oral function. Traditional stomatology relies on doctors' experience for diagnosis, which has problems such as strong subjectivity and low efficiency. In recent years, artificial intelligence has developed rapidly and has shown great potential in medical image analysis and disease diagnosis. Introducing artificial intelligence into stomatology is expected to break through the limitations of traditional diagnostic models and achieve accurate early screening of oral diseases.

[0003] In the prior art, some scholars have tried to use machine learning methods to perform intelligent analysis of oral images. Some scholars have proposed a dental caries X-ray diagnosis method based on support vector machines, which realizes automatic diagnosis of dental caries by extracting texture features and building a classification model. Some scholars have also proposed an intelligent recognition method for dental caries symptoms in intraoral color images based on convolutional neural networks. However, most of the existing oral image intelligent analysis methods are based on single-modality data and lack the comprehensive utilization of multi-dimensional medical information of patients. In addition, the machine learning model used is relatively simple and has limited feature representation capabilities, making it difficult to fully explore the inherent laws of oral diseases. Therefore, the present invention proposes an intelligent method that integrates oral multi-modality images and structured electronic medical records, integrates expert knowledge, and constructs an end-to-end deep learning diagnosis model, in order to obtain a more comprehensive and accurate diagnosis of oral diseases.

[0004] In view of the above problems, the present invention provides an early screening system for oral diseases based on an artificial intelligence-assisted diagnosis system. Summary of the invention

[0005] In view of the deficiencies in the prior art, the purpose of the present invention is to provide an artificial intelligence-based early screening method and system for oral diseases, which improves the efficiency and accuracy of oral disease diagnosis by integrating multimodal oral imaging data and electronic medical record information for intelligent modeling.

[0006] The specific technical solutions are as follows:

[0007] 1. An early screening method for oral diseases based on artificial intelligence, comprising the following steps:

[0008] (1) Using an oral image acquisition device to obtain oral multimodal image data of a patient, the multimodal image data includes an intraoral color image, a fluorescent image, and a spectral image; in this step, the intraoral color image is acquired by an intraoral color camera for subsequent visual analysis of the disease; the fluorescent image is acquired by an oral fluorescent imager, and the metabolic and structural abnormalities of oral tissues are identified through excitation and reception at specific wavelengths; the spectral image is acquired by a hyperspectral imager to obtain the spectral characteristics of different bands of oral tissues, and combined with subsequent spectral analysis methods, the precise positioning and identification of diseased tissues are achieved. The three types of oral image data to be acquired by the present invention complement each other and reflect the characteristics of oral diseases from different aspects.

[0009] (2) Collect the patient's electronic medical record information, including structured data such as oral-related symptoms and signs; in this step, the patient's electronic medical record information is collected through interviews and physical examinations, such as subjective symptoms such as pain, odor, and bleeding, other objective signs such as tooth mobility, gum color, and tartar index, as well as medical history data such as smoking history and oral hygiene habits. The collected information is standardized according to medical terminology to form a structured electronic medical record, which provides a clinical basis for subsequent fusion modeling. Structured electronic medical records can make full use of doctors' professional experience to systematically summarize the manifestations of the disease, which helps to improve the accuracy of diagnosis.

[0010] (3) Preprocess the collected multimodal image data, including image standardization, registration, fusion, etc., to obtain a unified feature representation; this step preprocesses the three types of image data separately. Among them, image standardization uses methods such as histogram equalization and white balance correction to achieve normalization of image brightness, contrast, and color, and eliminate interference from external factors such as external lighting; image registration uses operations such as feature point extraction and affine transformation to geometrically align different modal images to the same coordinate system, laying the foundation for pixel-level fusion; image fusion uses multi-scale fusion algorithms such as pyramid decomposition and wavelet transform to achieve the optimal combination of multi-source images in the frequency domain or spatial domain, thereby improving the information content and visual quality of the fused image. After standardization, registration and fusion, oral multimodal image data are integrated into a unified high-dimensional matrix representation, and each pixel combines multi-scale semantic information such as color, fluorescence, and spectrum, which is convenient for subsequent deep learning modeling.

[0011] (4) Using the pre-processed multimodal image features and combining them with electronic medical record information, an intelligent oral disease diagnosis model is constructed. This step uses deep learning technology to construct an end-to-end oral disease diagnosis model. First, the deep semantic features of the pre-processed multimodal image data are extracted. By designing a multi-channel convolutional neural network, the texture, morphology, color and other multi-dimensional feature patterns of the oral image are adaptively learned with different scales of receptive fields and convolution kernel combinations. At the same time, the electronic medical record information is converted into a structured semantic vector representation through One-Hot encoding, word embedding and other operations as an auxiliary input of the diagnosis model. Based on the deep feature extraction, a classifier network with an attention mechanism is built. By weighted fusion of multimodal image features and electronic medical record features, the most relevant information dimensions and data modalities for disease diagnosis are highlighted, thereby improving the classification and discrimination capabilities of oral diseases. The entire deep learning diagnosis model is jointly optimized in an end-to-end manner, with the disease type label as the supervision signal. The network parameters are adjusted through back propagation of the cross entropy loss function, and finally a nonlinear function is formed that maps oral multi-source heterogeneous data to disease classification results. The model is trained based on real patient data, using strategies such as data enhancement and transfer learning to improve the generalization performance of the model.

[0012] (5) Based on the trained intelligent diagnosis model, the newly collected patient images and medical records data are inferred and predicted, and the disease diagnosis results are output. For new patients, the collection and preprocessing of oral multimodal image data and electronic medical record information are completed according to the above process, and they are input into the trained deep learning diagnosis model, and the oral disease classification probability is obtained through the forward propagation calculation process. In addition to the identification of disease types, the diagnosis results also use visualization technologies such as heat maps to intuitively identify suspected lesions, estimate the scope and severity of lesions, and assist doctors in carrying out follow-up treatment. The present invention also supports doctors to interactively correct intelligent diagnosis results, continuously improve the knowledge base through human-computer collaboration, and form an application model in which intelligent diagnosis and expert experience complement each other.

[0013] 2. The present invention also provides an oral disease early screening system, comprising:

[0014] (1) Oral image acquisition device, used to acquire color, fluorescent and spectral images of the patient's oral cavity; this device consists of hardware devices such as an intraoral color camera, a fluorescent imager, and a hyperspectral imager, and acquires standardized patient oral image data through an automated acquisition process. During the acquisition process, the signal-to-noise ratio of the acquired image is improved through the optimization of photographic parameters and image enhancement, ensuring the quality of diagnosis while taking into account the patient's comfort. The acquisition device can be flexibly combined according to the needs of oral disease screening to achieve portable and desktop layouts.

[0015] (2) Electronic medical record collection module, used to collect structured data such as patient symptoms and signs; this module is connected to the hospital information system (HIS) and the clinician workstation (CWS), supporting doctors to directly enter information such as patient complaints and examination results into the system during daily consultations, and also supporting patients to self-assess symptoms and record physical signs through self-service terminals or mobile apps. The collected content is standardized according to medical terminology standards, disease diagnosis and treatment guidelines, etc., to ensure the standardization, integrity and interpretability of structured electronic medical records.

[0016] (3) Data preprocessing module, which is used to perform standardization, registration, fusion and other preprocessing on the collected multimodal images and structured medical record data; this module is based on a high-performance computing platform and uses medical image processing libraries such as OpenCV and ITK to achieve batch and automated preprocessing of oral multimodal images. Through the image standardization process, while maintaining the diagnostic semantics, non-uniform factors such as equipment differences and lighting changes are eliminated to achieve normalized representation of image data; through the image registration process, different modality images are mapped to the same spatial coordinate system to prepare for subsequent fusion; through the image fusion process, multi-source image information such as color, fluorescence, and spectrum are integrated at the pixel level to reveal the complex characteristics of oral diseases at the data level. This module is also responsible for preprocessing electronic medical record data such as cleaning and unified coding to match the image data format and provide standardized input for the intelligent diagnosis engine.

[0017] (4) Artificial intelligence diagnosis engine, which is used to integrate the pre-processed multimodal data to build a diagnosis model and perform intelligent diagnosis on new cases based on the model; the diagnosis engine is the core component of this system, and is embedded with an end-to-end oral disease diagnosis model based on deep learning. In the offline stage, the diagnosis engine obtains large-scale manually labeled data of oral diseases from channels such as hospital stock data and online public data sets to form a training set with disease classification as the label, and uses cross-validation, early stopping and other strategies to optimize the model's generalization ability and anti-overfitting ability. In the online stage, the diagnosis engine receives multimodal input from the data preprocessing module in real time, obtains disease classification results through forward propagation, presents them on the visual interactive terminal, and incorporates them into the remote consultation process. The diagnosis engine also has built-in model management and version iteration functions, and can continuously improve performance through active learning, incremental learning, etc., thereby giving the system the ability to evolve.

[0018] (5) Visual interactive terminal, used to show the diagnosis results to doctors and support doctors to revise and confirm the diagnosis results. The terminal adopts Web architecture design, has cross-platform and easy-to-expand features, and can be flexibly deployed on carriers such as doctor workstations and mobile ward rounds tablets, so that doctors can call auxiliary diagnosis services at any time. In terms of the presentation of diagnosis results, this terminal innovatively proposes heat map visualization, which intuitively represents the probability distribution of disease risk with color depth, and supports doctors to interactively adjust the heat map threshold to balance sensitivity and specificity. For the diagnosis results questioned by doctors, this terminal embeds an explainability module to explain the model reasoning basis from the perspectives of feature correlation and reference cases, thereby enhancing human-computer mutual trust. This terminal also provides human-computer collaborative functions such as annotation, comparison, and transmission to maximize the collaborative advantages of artificial intelligence diagnosis efficiency and expert experience.

[0019] (6) A remote consultation platform is used to submit difficult cases to dental specialists for consultation, and blockchain technology is used to ensure the secure sharing of patient privacy data during the remote consultation process. This platform is aimed at primary medical institutions and dental specialist hospitals, and uses the Internet to build a hierarchical diagnosis and treatment collaboration channel. This platform has a consultation application management function, supports primary doctors to initiate consultation needs online, matches the best expert resources through intelligent recommendations, and uses tools such as internal messages and video calls to achieve cross-hospital case discussions. During the consultation process, electronic medical records and medical images involving patient privacy need to be shared within the medical alliance. For this purpose, this platform adopts a blockchain solution, using asymmetric encryption, permission control, digital signature and other technologies to achieve reliable and controllable data flow throughout the process, and anti-tampering and anti-repudiation. The blockchain incorporates the nodes of each medical institution into a unified account book, and expands the rule engine through smart contracts to form a new remote consultation model with efficient collaboration and traceable supervision.

[0020] 3. The present invention provides an application process of an oral disease early screening system, comprising the following steps:

[0021] (1) The oral imaging device is used to collect intraoral color, fluorescence, and spectral images of the patient, and the electronic medical record collection module is used to record the patient's symptoms, signs and other structured data; this step is performed by the dentist, and the imaging device and the electronic medical record collection module should be pre-debugged and calibrated. The doctor needs to complete the standardized collection according to the system guidance and check the collection quality in a timely manner. The collection process should fully consider the patient's feelings, and explain the significance of the collection to the patient when necessary to eliminate concerns. For special groups such as the elderly and children, a personalized collection strategy should be adopted.

[0022] (2) The collected multimodal image data and structured medical record data are transmitted to the data preprocessing module, and after standardization, registration, fusion and other processing, a unified feature representation is formed, and then transferred to the artificial intelligence diagnosis engine; this step is automatically performed by the system, the transmission process adopts the DICOM medical imaging communication protocol, and the registration and fusion process adopts mature medical imaging processing library functions such as ITK and OpenCV. The processed multimodal data should meet the input format requirements of the intelligent diagnosis engine and can be automatically checked by the quality control tool.

[0023] (3) The AI ​​diagnosis engine automatically performs forward reasoning on new case data, outputs diagnosis results such as disease type, severity, and lesion location, and pushes them to the visual interactive terminal for display; this step is automatically called by the AI ​​diagnosis engine, and the reasoning calculation is performed on a high-performance server or cloud platform. The reasoning delay should be within an acceptable range. The diagnosis results should be formed into a standardized report, including the disease name, ICD code, severity score, thermal map annotation, etc. Identity authentication should be performed when pushing the diagnosis report to ensure data access security.

[0024] (4) The dentist views the intelligent diagnosis results through the visual interactive terminal, makes necessary corrections based on clinical experience and professional knowledge, and signs the opinion after the diagnosis is confirmed; this step is performed by the dentist on the visual interactive terminal. For the diagnosis results, the doctor can refer to the intelligent diagnosis explanation, and retrieve the patient's previous medical records, auxiliary examination data, etc. when necessary, and make a final diagnosis after comprehensive analysis. The doctor must be responsible for the diagnosis and has the right to reject the AI ​​diagnosis results. Identity authentication is required for signing the electronic signature, and the diagnosis opinion is included in the patient's electronic medical record management.

[0025] (5) For difficult cases that are still uncertain after AI diagnosis or the doctor's initial screening diagnosis, the doctor can initiate a provincial consultation through the remote consultation platform, and a senior doctor will make the final diagnosis; this step is completed by the dentist or specialist on the remote consultation platform. To initiate a consultation, you need to select a target specialist and upload relevant medical records. After accepting the invitation, the specialist can review the patient's medical records, and if necessary, remotely discuss with the dentist, have a video consultation with the patient, and give a consultation opinion on this basis. The entire process is supervised by the blockchain to ensure that data sharing and diagnosis and treatment behaviors are safe and reliable.

[0026] (6) Summarize the results of artificial intelligence and manual diagnosis, establish a screening diagnosis knowledge base, which is used to optimize the training of intelligent diagnosis models, and form patient diagnosis and treatment files to guide subsequent treatment. This step is automatically performed by the system. For a single case, the artificial intelligence diagnosis and the doctor's final diagnosis results are matched and aligned, and included in the long-term management of electronic medical records; for batch cases, the diagnosis results are summarized into structured knowledge, which is used to count the spectrum of oral diseases and form diagnosis and treatment standards. The system regularly retrains the intelligent diagnosis model, uses new case data for continuous learning, and corrects the erroneous cases reported by doctors to achieve closed-loop optimization of diagnostic effects.

[0027] Beneficial effects of the present invention:

[0028] 1. Innovatively integrate oral multimodal images and structured electronic medical records, covering more comprehensive diagnostic information dimensions than traditional methods, providing rich data support for accurate diagnosis of oral diseases.

[0029] 2. Use deep learning to build an end-to-end oral cavity disease diagnosis model. Through adaptive feature extraction and attention mechanism, fully explore the deep characteristics of diseases in multi-source heterogeneous medical data and significantly improve the disease diagnosis effect.

[0030] 3. Design heat map visualization and human-computer interaction functions to present AI diagnosis results in a way that is easy for clinicians to accept, enhance system usability, and effectively assist doctors in diagnosis and treatment decisions.

[0031] 4. Integrate remote consultation and blockchain technology, use artificial intelligence to achieve vertical allocation of oral medical resources, optimize the medical service system, and at the same time ensure that data sharing is safe and controllable, and promote the upgrade of tiered diagnosis and treatment.

[0032] 5. From data collection, model building to clinical application and continuous optimization, we provide a full-process intelligent solution for early screening of oral diseases, and realize an integrated closed loop of disease prevention, diagnosis and treatment.

[0033] In short, the present invention empowers oral medicine through artificial intelligence technology to create an innovative screening system that integrates image acquisition, multimodal fusion, intelligent diagnosis, and remote consultation. It provides a powerful tool for solving problems such as shortage of primary oral medical resources and uneven diagnostic levels. It has important practical significance and broad application prospects for improving people's oral health level and alleviating the pain points of people's livelihood such as difficulty in seeing a dentist. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic diagram of the overall structure of an oral disease early screening system based on an artificial intelligence-assisted diagnosis system to implement the present invention;

[0035] Figure 2A schematic diagram of an oral image acquisition device and an electronic medical record acquisition module for implementing the present invention;

[0036] Figure 3 A schematic diagram of a data preprocessing module for implementing the present invention;

[0037] Figure 4 A schematic diagram of an artificial intelligence diagnosis engine for implementing the present invention;

[0038] Figure 5 A schematic diagram of a visual interactive terminal for implementing the present invention;

[0039] Figure 6 A schematic diagram of a remote consultation platform for implementing the present invention;

[0040] Figure 7 A schematic diagram of an early screening process for oral diseases according to the present invention;

[0041] The following embodiments are only used to illustrate the principle and efficacy of the present invention, and are not intended to limit the present invention.

[0042] Example 1 A method for intelligent diagnosis of dental caries based on intraoral color images and electronic medical records

[0043] Step 1: Data collection. Use an intraoral color camera to collect images of the patient's oral surface, and record the patient's main symptoms, dietary hygiene habits and other electronic medical record information to establish an initial diagnostic data set. Invite three experienced dentists to annotate the image data for caries locations and grade the severity according to unified standards, which will serve as the gold standard for diagnostic learning.

[0044] Step 2: Data preprocessing. The histogram equalization algorithm is used to adaptively adjust the image brightness and contrast, the white balance correction is used to suppress color cast, and the median filter is used to remove high-frequency noise and retain the typical texture features of caries. The image is registered to a unified coordinate system according to the shooting position and angle. The structured information of the electronic medical record is extracted and the text is mapped into a semantic vector.

[0045] Step 3: Feature extraction. For color images, the pre-trained ResNet-101 deep convolutional neural network is used to extract high-level semantic features of the caries area. Two scale receptive field convolution kernels of 8×8 and 16×16 are set to focus on the local and overall morphological changes of caries, respectively. Finally, the two scale feature maps are fused to obtain a 1024-dimensional feature vector. For the electronic medical record semantic vector, the bidirectional long short-term memory network BLSTM is used to learn the medical record correlation and temporal features.

[0046] Step 4: Diagnostic modeling. Design a fully connected classifier network with an attention mechanism, and adaptively fuse image and medical record features according to the attention weight to achieve caries classification and severity regression. Weighted Focal Loss and L1 Loss are used to jointly optimize the classification and regression tasks for positive sample data. The AdamW optimizer fine-tunes the pre-trained network parameters, with a batch size of 16, an initial learning rate of 1e-4, and a maximum number of iterations of 10,000.

[0047] Step 5: Model evaluation. Using 5-fold cross validation, the accuracy, precision, recall, and F1 values ​​were used to evaluate the classification performance of the model, and the mean absolute error (MAE) and root mean square error (RMSE) were used to evaluate the regression performance. On the test set, the classification task finally achieved an accuracy of 92%, a precision of 90%, a recall of 94%, and an F1 value of 92%; the regression task MAE was 0.12 and the RMSE was 0.18. The recall rate reflects the ability of this method to screen people at high risk of dental caries, which is of great significance for early intervention treatment.

[0048] Step 6: Heat map visualization. For new diagnostic images, extract image features and input them into the trained model, calculate the category probability score of each pixel, generate a heat map and superimpose it on the original image to intuitively display the suspected caries area. Render blue, yellow, and red according to the probability threshold to represent mild, moderate, and severe caries risks.

[0049] Step 7: Human-computer collaborative diagnosis. The original images, heat maps and diagnostic reports are displayed to doctors through the Web system for their reference. Doctors can interactively adjust the heat map threshold and manually circle the lesions to obtain satisfactory diagnostic results. The diagnostic report and doctor feedback are automatically returned to the training data set to update the iterative model.

[0050] This method is applied in routine diagnosis and treatment to conduct rapid intelligent screening of newly diagnosed patients, provide auxiliary suggestions for doctors to diagnose caries, reduce workload and improve screening efficiency. In future community free clinics and other scenarios, patients can upload oral images by themselves, and remote intelligent diagnosis can be used to preliminarily determine the risk of caries, and timely warn medical treatment, which is expected to significantly reduce the proportion of missed caries diagnosis and delayed treatment. Example 2 A method for intelligent screening of oral mucosal precancerous lesions based on spectral and fluorescence imaging

[0051] Step 1: Data collection. Use a hyperspectral camera to collect spectral data of oral mucosal tissues and identify the spectral difference characteristics of different pathological types of tissues; use a fluorescence imaging system with wide-spectrum excitation and narrow-band emission to capture the autofluorescence and induced fluorescence images of diseased tissues, which sensitively reflect changes in tissue metabolic status. Label the types of precancerous lesions such as leukoplakia, erythema, lichen planus, and submucosal fibrosis, and establish a screening data set.

[0052] Step 2: Spectral analysis. The collected hyperspectral data is preprocessed by scattering correction and fluorescence background subtraction to extract typical spectral features such as absorption spectrum and excitation emission valley peak. Dimensionality reduction is performed using principal component analysis (PCA) and independent component analysis (ICA) to obtain characteristic spectra of different pathological tissues. Support vector machine (SVM) is used to classify characteristic spectra and distinguish precancerous lesions from normal tissues.

[0053] Step 3: Fluorescence analysis. Normalize the collected autofluorescence images and use the excitation spectrum to correct the intensity distribution of the induced fluorescence images. Extract the morphological parameters of the lesion area and analyze the spatial heterogeneity of the fluorescence intensity. For suspicious lesion tissues, extract the multi-band fluorescence signal intensity ratio to quantitatively describe the degree of metabolic abnormality of the lesion. Use the random forest algorithm to classify the fluorescence features and identify the type of precancerous lesions.

[0054] Step 4: Model fusion. Construct a deep learning network with two parallel paths, spectral and fluorescence, and fuse the diagnostic results of the two modalities at the decision layer to improve the detection rate of oral mucosal lesions. Use the cross entropy loss function and SGD optimizer to train the model, with the regularization parameter set to 0.01, the dropout probability to 0.5, and the maximum number of iterations to 5000.

[0055] Step 5: Performance evaluation. The diagnostic results were compared with the pathological gold standard, and the receiver operating characteristic curve (ROC) and area under the curve (AUC) were used to evaluate the diagnostic performance of the model. In the test set, the AUC of the spectral and fluorescence fusion model for oral mucosal precancerous lesions reached 0.93, which was 8% and 11% higher than that of the single-modality model, respectively. For early lesions such as squamous cell carcinoma, the sensitivity can reach more than 85%, which is expected to reduce the risk of missed diagnosis.

[0056] Step 6: 3D reconstruction and visualization. Align the spectrum and fluorescence images to the same spatial coordinate system, outline the lesion tissue in 3D, and form a 3D lesion model. Render the spectrum and fluorescence signal intensity of the lesion area to intuitively reflect the metabolic abnormality information of the lesion site. The 3D reconstruction results can be rotated 360 degrees and enlarged to facilitate all-round observation by doctors.

[0057] Step 7: Design of clinical screening pathways. Develop spectral and fluorescence screening programs for high-risk groups such as smokers, clarify the collection process and diagnostic criteria. Conduct community screening regularly, establish a database of oral mucosal precancerous lesions, follow up suspected cases, and optimize personalized intervention measures. Screening data is used to continuously optimize the intelligent diagnosis model.

[0058] This method combines spectral and fluorescence features, giving full play to the qualitative diagnostic advantages of hyperspectral imaging and the quantitative analysis advantages of fluorescence imaging, and significantly improves the detection and identification of precancerous lesions compared with traditional methods. In clinical promotion, spectral and fluorescence acquisition devices can be integrated into portable devices such as smart endoscopes to achieve non-invasive, real-time, in situ diagnosis, and assist in early screening of primary precancerous lesions. In subsequent treatment, a three-dimensional model of the lesion can also be used to formulate a personalized surgical plan to guide precise resection.

[0059] Example 3: A remote intelligent diagnosis platform for oral X-ray images for community hospitals

[0060] Module 1: Remote diagnosis service. Build a remote diagnosis platform with B / S architecture to support doctors in community hospitals to upload patients' intraoral X-rays, panoramic films, and CBCT images to the cloud through a web browser, which will be automatically analyzed by the artificial intelligence diagnosis engine. The diagnosis content includes caries, periapical lesions, alveolar bone absorption, etc. Blockchain technology is introduced to store medical data on the chain to ensure the security of remote transmission.

[0061] Module 2: Expert consultation network. For difficult cases that are uncertain in intelligent diagnosis, community doctors can initiate remote consultations, and oral radiology experts from provincial hospitals will participate in the diagnosis. Experts communicate with community doctors through in-site communication tools, review patients' previous imaging data, and give diagnostic opinions after comprehensive judgment. The consultation opinions use expert electronic signatures, are embedded in the imaging data package, and synchronize the diagnosis and treatment records of all parties.

[0062] Module 3: Mobile diagnosis and treatment app. Develop a mobile application for doctor-patient interaction. Patients can make appointments for examinations, view diagnostic reports and medical reminders. Doctors can work on the go, review intelligent diagnostic results in a timely manner, and conveniently manage patient follow-up. The app also supports online consultation between doctors and patients, analyzes patients' oral images and self-reported symptoms, and guides reasonable medical treatment. Asymmetric encryption is used for data transmission to protect patient privacy.

[0063] Module 4: Diagnostic knowledge learning. To ensure the quality of intelligent diagnosis, an oral imaging expert diagnosis knowledge base is established, and provincial and municipal experts are regularly organized to annotate cases and accumulate gold standard data sets. At the same time, an active learning model is introduced, and the intelligent diagnosis model selects the most valuable samples for diagnosis from a large number of cases for experts to revise and include in training. The knowledge iteration cycle can be optimized to once a month.

[0064] Module 5: Quality Control and Audit. Design medical quality management standards, monitor community doctor applications and expert diagnosis behaviors in real time, and conduct batch review of intelligent diagnosis results. With the help of natural language processing, analyze the standardization of diagnostic report words and sentences, count the time spent in each stage of diagnosis, and control the quality of medical services throughout the entire process. Quality control records are stored on the chain to facilitate subsequent audits by medical authorities.

[0065] Deployment plan: Build an intelligent diagnosis engine and knowledge management system on the provincial hospital radiology cloud platform, connect to the hospital PACS to achieve data interaction; deploy a remote image upload gateway in the community health service center, and integrate the diagnosis platform call interface. Doctors and patients use the oral health all-in-one machine to collect oral images by themselves, or community doctors upload them manually. The remote diagnosis report is pushed to the community doctor workstation and synchronized to the cloud electronic medical record and patient app.

[0066] Typical application scenario: A suspected positive case was detected by an oral X-ray camera at a community hospital in the eastern province of S, and the local doctor uploaded it to the intelligent diagnosis platform. The diagnosis engine automatically screens for multiple oral diseases and provides feedback on the diagnosis results within 10 minutes, with a preliminary diagnosis of apical periodontitis of the lower left posterior teeth. As the image signal of the lesion is not clear, the system recommends submitting a remote consultation. The community doctor initiates a consultation application on the platform, and the director of the radiology department of the provincial oral hospital interprets the image within 24 hours, uses graphic annotations to circle the scope of the lesion, and reports a clear diagnosis of chronic apical periodontitis, recommending root canal treatment. The consultation report is simultaneously pushed to the community doctor workstation and the patient app to guide the patient to seek medical treatment in a timely manner and inform the precautions. The patient follows the doctor's advice to the oral department of the superior hospital for treatment, makes an appointment for a CBCT review, further evaluates the scope of the lesion, and formulates a root canal treatment plan. After treatment, the imaging data is automatically returned to the community for follow-up of the efficacy.

[0067] Example 4 A three-dimensional intelligent design method for oral restorations

[0068] Step 1: Data collection. Use a 3D scanner dedicated to oral restoration to scan the patient's oral model, including tooth preparations, opposing teeth, jaw bone relationships, etc., and reconstruct a 3D digital model of the oral cavity. Standardize the scanning process, control the scanning accuracy to better than 0.01mm, and ensure data integrity. Input design requirements for restoration material parameters, mechanical properties, aesthetic requirements, etc.

[0069] Step 2: Intelligent restoration design. Import the digital oral model into the intelligent restoration design platform, and automatically generate the initial shape of the restoration through computer-aided design. The platform has a built-in database of 1,000 standard tooth shapes for reference; at the same time, based on artificial intelligence algorithms, it automatically matches the optimal initial design plan based on the patient's oral morphological characteristics, saving design time.

[0070] Step 3: Oral function simulation optimization. Based on the initial restoration morphology, digital simulation analysis of oral functions such as occlusion and restoration stress was carried out. The finite element analysis method was used to evaluate the distribution of chewing stress on the restoration, and 1×106 cycles of loading were simulated to determine the risk of fatigue failure. The inverse optimization algorithm was used to automatically adjust the restoration morphology under the guidance of the stress cloud map to obtain the restoration plan with the best mechanical properties.

[0071] Step 4: Aesthetic parameter adjustment. Optimize the aesthetic parameters of the restoration, such as the contour curve, color and texture. The platform provides common tooth morphology aesthetic parameter templates, which can be flexibly adjusted through interactive parameter sliders, such as crown aspect ratio, cusp bevel angle, dentin hue saturation, etc. The platform previews the restoration's three-dimensional morphology and true color rendering effects in real time, assisting doctors and patients in evaluating the aesthetic effect and finding the best balance between function and aesthetics.

[0072] Step 5: Virtual restoration design verification. The optimized 3D restoration model is virtually tried on with the patient's original oral model to check the restoration's positioning, edge tightness, and distribution of adjacent contact points, and eliminate interference factors. Simulate the midline, smile curve, and tooth-lip relationship after restoration to evaluate the overall aesthetic effect of the oral cavity. For problems found during the trial, return to the optimization process and continue to adjust until a restoration plan that satisfies the doctor and patient is obtained.

[0073] Step 6: Processing-assisted design. According to the material characteristics of the restoration, machine processing-assisted optimization is performed. For hard and brittle materials such as zirconia porcelain, the intelligent platform automatically adds the restoration edge width, calculates the optimal molding blank size, and plans refined processing paths and tool parameters to minimize material removal and surface defect risks. For PEEK high-performance composite resin, the injection molding process parameters are optimized, and injection pressure and temperature changes are simulated to avoid molding shrinkage and warping deformation.

[0074] Step 7: Personalized repair service management. Share patient repair needs and design plans through the Internet cloud platform, and manage the design process online. Patients can check the iteration of the repair plan at any time and provide feedback; doctors can grasp the processing progress in real time and optimize logistics distribution. All data is encrypted and managed through blockchain technology, which is traceable and cannot be tampered with, ensuring patient privacy and security. Subsequent repair data is linked to the patient's lifelong electronic medical record for follow-up visits.

[0075] This method makes full use of artificial intelligence parametric design and simulation optimization technology to digitize and automate the traditional oral restoration process, greatly improving the efficiency of restoration design and quality stability. Expert experience is solidified on the design platform through algorithms, promoting the sinking of high-quality medical resources and promoting the precise and personalized development of oral restoration. Patients can enjoy high-quality restoration customization services and improve their medical experience.

[0076] Example 5: A periodontitis intelligent diagnosis and treatment decision support system

[0077] Module 1: Multimodal data collection. Through the Internet of Things technology, the patient's structured electronic medical records, intraoral color images, full-mouth curved surface CBCT images, microbial gene sequencing data, etc. are collected and integrated into the periodontitis intelligent auxiliary diagnosis and treatment platform to form a comprehensive oral health record for the patient. Various types of data are semantically mapped and spatially registered through artificial intelligence algorithms to achieve multimodal data fusion and correlation analysis.

[0078] Module 2: Intelligent diagnosis of periodontitis. Design targeted deep learning models for feature extraction and representation learning of multi-source heterogeneous medical data. For example, use the Mask R-CNN algorithm to segment the alveolar bone contour in CBCT images and quantify bone resorption; use transfer learning and fine-tuning strategies to optimize the image segmentation network and adapt to the differences in bone density distribution under different imaging devices; design a periodontitis grading discriminator to analyze clinical manifestations such as gingival redness, bleeding, and tartar, and evaluate the severity of inflammation; construct a host-microbe interaction network to reveal the association between periodontal pathogenic bacteria and the body's immune factors, and predict disease activity. The diagnostic results are combined with the patient's main symptoms and risk factor analysis to characterize the severity and prognosis of periodontitis in multiple dimensions.

[0079] Module 3: Intelligent decision-making on treatment plans. Based on the diagnostic classification of periodontitis, similar cases in the case database are recalled, and successful treatment experience is learned. The decision tree algorithm automatically generates personalized treatment plans. Combined with factors such as the patient's oral hygiene habits and compliance with medical treatment, a long-term care plan is formulated to conduct disease risk assessment. The treatment plan covers a combination of various means such as medication, cleaning procedures, restoration design, and laser treatment, striving for systematic radical cure. For difficult cases, the platform is connected to the artificial intelligence-assisted consultation system, and provincial oral specialists are invited to review online, provide guidance on the treatment plan, and ensure the standardization of the plan.

[0080] Module 4: Intelligent management of the rehabilitation process. After surgery, patients use a "smart toothbrush" to clean their mouths, collect data such as brushing time and strength, and upload them to the rehabilitation management system; at the same time, embedded sensors can detect oral health indicators such as saliva Ph value and bleeding index to evaluate disease recovery. The system provides personalized diet and exercise education based on the patient's rehabilitation progress, and customizes the frequency of follow-up visits and examination items. At the same time, it pushes warnings for high-risk patients to doctors, guides correct care through remote video, and intervenes in possible complications as soon as possible. All diagnosis and treatment data are automatically written to the blockchain to facilitate medical quality control and doctor-patient rights protection.

[0081] Module 5: Disease prevention population monitoring. Taking advantage of artificial intelligence, intelligent analysis of community residents' oral health records was conducted to identify high-risk groups for periodontitis; targeted inspection plans were customized for high-risk groups, and door-to-door oral hygiene guidance was provided. Medical insurance data was integrated to analyze the medical treatment trends of patients in different regions, and periodontitis disease maps were drawn to guide the allocation of oral public health resources; through big data correlation analysis, the causal relationship between periodontitis and systemic diseases was studied to provide evidence-based basis for the formulation of oral health promotion policies. All monitoring data were anonymized and shared for clinical research use.

[0082] This system covers the whole life cycle management of periodontitis prevention, diagnosis, treatment and rehabilitation, takes advantage of artificial intelligence technology, integrates radiomics, case database, big data monitoring and other means to achieve intelligent and refined management. In clinical diagnosis and treatment, the system can assist doctors in formulating reasonable diagnosis and treatment plans, standardize diagnosis and treatment behaviors, and improve the overall medical quality; remote consultation can break the bottleneck of scarce specialist resources and promote hierarchical diagnosis and treatment; in community prevention and control, the system can assist in oral health big data monitoring, timely discover high-risk factors for diseases, and provide intellectual support for precise prevention. Patients can obtain convenient and high-quality diagnosis and treatment services and improve their medical experience.

[0083] Example 6: A grass-roots-oriented children's oral health artificial intelligence system

[0084] Module 1: Caries risk assessment. Information on children's eating habits, oral hygiene behaviors, etc. was collected through questionnaires, and saliva samples were collected to detect the level of cariogenic bacteria. The risk factors were structured and input into the caries risk assessment model. The logistic regression algorithm was used to model the probability of caries risk for individual children. The risk level was weighted with weight factors such as the child's age and family history to develop an oral health management plan.

[0085] Module 2: Oral health education. Develop targeted oral health knowledge learning courseware based on the cognitive level of children of different ages. Use artificial intelligence education technology to achieve networked and interactive learning. For example, in early childhood, the focus is on developing brushing habits. The platform integrates parent-child games and stimulates children's interest in active learning through virtual pets and other forms; in school age, focus on popularizing oral health knowledge, increase knowledge questions and answers and scenario simulations, and deepen the understanding of dental caries prevention; in adolescence, emphasize responsibility awareness, conduct online knowledge competitions, and guide children to master self-oral management skills. All learning processes are digitized to facilitate timely intervention and guidance by parents and doctors.

[0086] Module 3: Intelligent planning of pit and fissure sealants. Artificial intelligence algorithms are used to analyze the morphological characteristics of children's oral digital models and X-ray images, and to assess the susceptibility to caries. Through big data mining of caries-prone sites and tooth surfaces, combined with individual salivary cariogenic bacteria level detection, the number and location of permanent teeth that need preventive sealing are determined. The platform provides the best sealing time prompt based on the patient's age and the eruption of the opposing teeth. Patients can intuitively understand the necessity of treatment through the mobile terminal, and doctors can use the system to standardize operations to ensure the quality of medical care.

[0087] Module 4: Oral growth and development monitoring. Build a digital database of children's oral archives and collect oral 3D imaging data. Automatically identify features and measure parameters of the digital jaw model, including bone face shape, dental arch morphology, and the order and time of permanent tooth eruption. Use clustering and association analysis to identify the risk of jaw development abnormalities and predict the possibility of malocclusion. Combined with skull measurements and facial photos, evaluate the overall growth pattern of the child's craniofacial region. The large database supports doctors in formulating personalized correction plans, dynamically monitoring the correction process, and evaluating long-term efficacy.

[0088] Module 5: Children's Oral Health Management. Through the three-in-one oral health network of hospitals, schools, and communities, oral health management is implemented for children in the jurisdiction. Regular oral examinations are organized for school-age and school-age children, oral imaging data are collected, and oral diseases are screened by artificial intelligence; follow-up visits are conducted for children at medium and high risks, personalized prevention and treatment plans are formulated, and follow-up visits are flexibly adjusted; schools and communities are organized to carry out oral health day education, popularize oral health knowledge, and mobilize parents to participate in children's oral management. Integrate medical insurance and electronic health record data, analyze the factors affecting children's oral health, and provide data support for the formulation of children's oral health policies.

[0089] The system is based on the needs of primary medical preventive health care, integrating artificial intelligence technology with the existing children's oral health care model to form an artificial intelligence-enabled closed loop of education and publicity-intelligent screening-personalized intervention-dynamic management. Through the collection and analysis of the panoramic data of children's oral health in the community, early detection of oral diseases and abnormal risks will help to carry out precise prevention; personalized education will help children establish correct oral health concepts and develop good hygiene habits; standardized intelligent management networks will help improve the quality of oral health care and service efficiency. In short, the application of this system will help cultivate children's oral health awareness from an early age and promote the healthy and happy growth of children.

[0090] Example 7 An artificial intelligence-assisted orthodontic system

[0091] Module 1: Intelligent correction design. Use structured light scanning to obtain a digital impression of the patient's oral cavity, replacing the traditional physical impression-making process. Apply artificial intelligence algorithms to standardize digital impressions, and use three-dimensional data registration technology to fuse digital impressions with cephalometric data and facial photos to obtain refined craniofacial three-dimensional morphological information. Relying on the orthodontic expert knowledge base, an artificial intelligence model for deformity diagnosis and correction design is established. According to the specific conditions of the patient, such as the degree of dentition crowding, dental arch morphology, occlusal relationship, etc., the deformity type is automatically diagnosed, and the jaw facial morphology after orthodontic treatment is simulated to develop a personalized correction plan. The design plan fully considers aesthetic functional factors such as facial coordination, smile curve, and tooth-lip relationship, while taking into account the feasibility of correction to obtain satisfactory treatment results.

[0092] Module 2: Three-dimensional design and production of orthodontic devices. Based on the correction design plan, digital technologies such as CAD / CAM and 3D printing are used to realize the intelligent design and production of orthodontic devices. For fixed orthodontic devices, the system can optimize the position and torque control of brackets, accurately bond, and produce personalized alloy wire arches, giving full play to the mechanical effect of instrument correction; for invisible orthodontic devices, the system can optimize the three-dimensional structural parameters of the orthodontic device, digitally simulate the tooth movement process, reasonably arrange the number of visits and wearing cycles, and generate three-dimensional animations to guide patients to use the orthodontic device to improve compliance. All design data are printed and formed by a 3D printer to form a personalized device that is highly matched with the patient's oral anatomy. The device production process is digitalized to facilitate doctor-patient communication and quality traceability.

[0093] Module 3: AI-assisted bonding navigation. Using the patient's digital intraoral impression obtained by oral scanning and the ideal bracket position in the correction design plan, the best bonding path is formulated through virtual intraoral environment matching. Augmented reality navigation is used through visualization technology to guide doctors to accurately bond brackets. During the navigation process, the deviation between the doctor's actual operation position and the ideal position is monitored, and the deviation data is fed back in real time for continuous learning and optimization. This module integrates AI perception, modeling, control and other skills, realizes "AI+doctor" collaboration, and improves the accuracy and quality controllability of orthodontic treatment. Patient bonding data is stored in personal files for subsequent efficacy evaluation.

[0094] Module 4: Intelligent efficacy monitoring and prognosis evaluation. During the patient's treatment, oral scans are used to obtain three-dimensional oral images at different stages, which are automatically registered and compared with baseline images to quantitatively evaluate tooth movement, analyze the deviation between the treatment process and expectations, predict the long-term treatment effect, and provide a reference for program adjustment. At the same time, the patient's wearing experience and facial feeling data are dynamically collected to comprehensively evaluate the treatment comfort and facial shape improvement. For cases suspected of poor comfort and unsatisfactory progress, the system automatically issues an early warning to prompt the doctor to intervene. The treatment efficacy data is continuously accumulated and can be used for continuous learning of artificial intelligence models to improve the interpretability and success rate of design solutions. Patients can view the progress of treatment in real time through mobile terminals, enhance their confidence, and cooperate with doctors to complete treatment.

[0095] Module 5: Remote orthodontic service platform. In view of the current situation of lack of professional orthodontists in remote areas, a remote orthodontic service network platform covering the county, town and village levels is established. Through the wearable intelligent oral image acquisition device, patients can complete the oral image acquisition by themselves and upload other medical history data to the cloud; county-level dental hospitals use artificial intelligence-assisted systems for initial screening and design, and submit difficult cases to provincial hospitals for consultation; provincial-level specialized hospitals organize multidisciplinary remote consultations to review correction design plans. The hierarchical diagnosis and treatment model makes full use of the expert resources of superior hospitals and the service network of lower-level hospitals to improve the efficiency of medical resource allocation. At the same time, the remote platform is embedded with an artificial intelligence quality control system to standardize and refine the management of the entire diagnosis and treatment process of each case to ensure the quality of medical services.

[0096] The present invention is aimed at a new generation of intelligent orthodontic diagnosis and treatment mode. It is based on the patient's three-dimensional image and facial data, with artificial intelligence algorithms as the core and 3D printing as the means to build a full-process digital platform covering diagnosis, design, treatment and monitoring. Compared with the traditional diagnosis and treatment path, the platform can significantly improve the efficiency of orthodontic plan formulation, optimize the device design and production process, realize the visualization and quantifiable management of the diagnosis and treatment process, and thus improve the correction effect and medical experience. At the same time, the platform makes full use of information technology such as the Internet and big data to explore new remote orthodontic models adapted to the grassroots level, promote the sinking of high-quality medical resources, and allow more patients in remote areas to enjoy high-quality oral orthodontic services, reduce academic interruptions and family burdens caused by orthodontic treatment, and improve the quality of life.

Claims

1. An early screening method for oral diseases based on an artificial intelligence-assisted diagnosis system, comprising: using an oral image acquisition device to obtain a patient's oral multimodal image data, the multimodal image data including an intraoral color image, a fluorescent image, and a spectral image; acquiring the patient's electronic medical record information, including structured data such as oral-related symptoms and signs; preprocessing the acquired multimodal image data, including image standardization, registration, fusion, etc., to obtain a unified feature representation; using the preprocessed multimodal image features in combination with the electronic medical record information to construct an oral disease intelligent diagnosis model; based on the trained intelligent diagnosis model, inferring and predicting the newly acquired patient images and medical record data, and outputting the disease diagnosis results.

2. The method according to claim 1, characterized in that ,The oral image acquisition equipment includes an intraoral color camera or an intraoral camera, a ,fluorescence imager and a hyperspectral imager.

3. The method according to claim 1, characterized in that ,The image preprocessing includes image standardization, registration and fusion, and the ,pixel-level fusion of multimodal images is achieved by using algorithms such ,as image pyramid or wavelet transform.

4. The method according to claim 1, characterized in that ,The intelligent diagnosis model for oral diseases is constructed using deep learning ,algorithms, including convolutional neural networks, graph convolutional networks, long short ,time memory networks or transformer networks, and the joint modeling of ,multimodal imaging features and structured medical record features is achieved through ,end-to-end training.

5. The method according to claim 1, characterized in that ,The disease diagnosis results include multi-dimensional information such as disease classification, lesion location, and severity, which are presented in the form of a heat map, and the suspected lesion areas are marked with different color brightness according to the disease type and severity.

6. An oral disease early screening system based on an artificial intelligence-assisted diagnosis system, comprising: an oral image acquisition device for acquiring color, fluorescent and spectral images of the patient's oral cavity; an electronic medical record acquisition module for acquiring structured data such as symptoms and signs of the patient; Data preprocessing module, used to perform standardization, registration, fusion and other preprocessing on the collected multimodal images and structured medical record data; The artificial intelligence diagnosis engine is used to integrate pre-processed multimodal data to build a diagnosis model, and to perform intelligent diagnosis of new cases based on the model; the visual interactive terminal is used to display disease diagnosis results to doctors and support doctors to correct and confirm the diagnosis results.

7. The system according to claim 6, characterized in that ,The artificial intelligence diagnosis engine adopts deep learning algorithms such as convolutional neural networks, graph convolutional networks, long short term memory networks or transformer networks to realize multimodal data fusion modeling, and integrates manually annotated disease type labels for end-to-end training.

8. The system according to claim 6, characterized in that ,The visual interactive terminal presents the disease diagnosis results in the form of a heat map, ,colors the suspected lesion areas, and supports the doctor to manually adjust the ,heat map threshold to optimize the diagnosis effect.

9. The system according to claim 6, characterized in that , also includes a remote consultation platform for submitting difficult cases to dental specialists for consultation, and using blockchain technology to ensure the secure sharing of patients' privacy data during remote consultations.

10. The working method of the system according to claim 9, characterized in that , including: using the oral image acquisition device and the electronic medical record acquisition module to obtain the patient's multimodal image data and structured medical record data; using the data preprocessing module to perform standardization, registration, fusion and other preprocessing on the collected multimodal images and medical record data; inputting the preprocessed data into the artificial intelligence diagnosis engine, fusing the multimodal data to build a diagnosis model; using the trained diagnosis model to perform intelligent diagnosis on the image and medical record data of new cases and generate a disease diagnosis report; presenting the diagnosis results to the doctor through a visual interactive terminal, and supporting the doctor to correct and confirm the diagnosis results; for difficult cases with uncertain diagnosis, submit them to oral specialists through the remote consultation platform to obtain consultation opinions and assist primary doctors in making diagnostic decisions.

Images