Oral ceramic restoration optical prediction matching method and system based on deep learning

Abstract

The invention provides an oral ceramic restoration optical prediction matching method and system based on deep learning, relates to the technical field of oral medicine, and solves the limitation problems that existing oral restoration is low in clinical colorimetric accuracy, depends on artificial experience, has medical technology communication information deviation and the like. The method comprises the following steps: acquiring spectral data of natural teeth, dentin, enamel and different types of ceramics, and calibrating the semi-permeability of enamel and ceramic samples by using an extinction coefficient; obtaining a ceramic-dentin complex and composite spectrum data thereof, constructing a multi-layer sensor model and completing training; and splicing the dentin spectral data and the ceramic spectral data into an initial input vector, splicing the initial input vector with the calibrated semi-permeable constraint vector, and inputting the spliced initial input vector into a multi-layer sensor model to obtain a model matching result. The intelligent model for automatically generating the natural tooth defect repair scheme is established based on the deep learning technology, the intelligent repair design and the visual repair effect are achieved, and clinical application is effectively guided.

Description

Technical Field

[0001] The present invention relates to the technical field of oral medicine and is applied in the process of oral restoration treatment, and specifically to an optical prediction and matching method and system for oral ceramic restorations based on deep learning. Background Art

[0002] In the field of oral restoration treatment, the repair of tooth defects not only needs to rebuild the chewing function of teeth, but also needs to achieve an aesthetic effect that is coordinated with natural tooth tissue. Natural teeth have a unique composite layered structure, and their enamel layer, dentin layer and junction area show gradually changing translucent characteristics and three-dimensional layering in optical properties. Traditional restoration production mainly relies on visual colorimetry to determine tooth color, but this method is significantly affected by ambient light intensity, observation angle and the operator's color discrimination ability, and is prone to subjective judgment bias. Although the dental colorimeter improves the objectivity of color measurement through spectral analysis, its single-point sampling technical characteristics make it difficult to fully reflect the three-dimensional distribution law of tooth surface color, and color space conversion distortion is prone to occur when the measurement data is transmitted to the restoration production link.

[0003] With the advancement of digital technology, equipment such as spectrophotometers and intraoral scanners are gradually used in clinical practice to obtain tooth morphology and color information through spectral imaging and three-dimensional morphology reconstruction technology. However, the existing technology for analyzing the multi-layer optical properties of teeth is still limited to the planar collection of surface colors, and has failed to establish a quantitative correlation model between the light transmittance characteristics of enamel, the scattering effect of dentin, and the optical parameters of restorative materials. Clinical observations have shown that the differences between restorations and natural teeth in key optical indicators such as light transmittance and fluorescence response often lead to non-physiological reflections or shadows on the restoration interface under natural light. The essence of this is that the traditional colorimetric system only focuses on the numerical matching of the two-dimensional color space, while ignoring the three-dimensional regulation mechanism of the tooth structure on the light propagation path.

[0004] In the field of materials science, the Kubelka-Munk theory has been introduced into the study of layered color matching of dental resins, but there is a mismatch between its theoretical assumptions and the actual structural characteristics of oral restorations. Especially when the thickness of the restoration is significantly reduced, the color deviation between the predicted results of the traditional model and the actual optical performance is easily discernible to the naked eye.

[0005] Therefore, the intelligent transformation of current oral rehabilitation technology faces dual bottlenecks: First, there is information attenuation in the digital impression colorimetric data during the interactive transmission between medical technicians; second, the application of machine learning technology in dental color matching mostly focuses on two-dimensional image processing, and a three-dimensional prediction model has not been constructed by combining the spatial optical characteristics of tooth microstructure. Although hyperspectral imaging technology can obtain multi-dimensional optical information of teeth in a wide spectral range, the complexity of its data processing and the timeliness requirements of clinical operations have not yet reached an effective balance, restricting the practical application of this technology in immediate restoration scenarios. Summary of the Invention

[0006] The object of the present invention is to solve the problems of low accuracy in current oral rehabilitation clinical color matching, the dependence of the restoration effect on the subjective judgment of the human eye, and the information deviation in medical technician communication due to low technological intelligence. Therefore, an optical prediction matching method and system for oral ceramic restorations based on deep learning are proposed. The present invention analyzes the optical characteristics of natural teeth and ceramic restoration materials, can establish an optical database with a corresponding mapping relationship, and based on deep learning technology, establishes an intelligent model for automatically generating natural tooth defect restoration plans, realizing the intelligence of restoration design and the visualization of restoration effects, and can well guide clinical applications.

[0007] The present invention adopts the following technical solutions to achieve the object:

[0008] An optical prediction matching method for oral ceramic restorations based on deep learning, comprising the following steps:

[0009] S1. Obtain the spectral data of natural teeth, then separate the dentin and enamel of the natural teeth, and obtain the dentin spectral data and enamel spectral data respectively; obtain the spectral data of different types of ceramics;

[0010] S2. Use the extinction coefficient to calibrate the semi-permeability of enamel and the semi-permeability of the ceramic samples corresponding to the ceramic spectral data, and combine the calibration results of the two to form a semi-permeability constraint vector;

[0011] S3. Bond the ceramic samples to the separated dentin to obtain a ceramic-dentin composite; obtain the composite spectral data corresponding to the ceramic-dentin composite;

[0012] S4. Construct a multi-layer perceptron model, and based on all types of spectral data obtained, perform model training; then splice the dentin spectral data and the ceramic spectral data to form an initial input vector;

[0013] S5. Splice the initial input vector and the semi-permeability constraint vector to form an extended input vector, and input the extended input vector into the trained multi-layer perceptron model to obtain a model matching result;

[0014] In step S5, the trained multi-layer perceptron model performs optical prediction matching on the composite spectrum data of the ceramic-dentin complex and the natural tooth spectrum data, and uses the calibrated semi-permeability of the enamel and ceramic samples as a constraint condition to obtain the type of ceramic sample that meets the constraint condition, and the output is the model matching result.

[0015] Preferably, it further includes the steps: S6. Establish an optical database on the mapping relationship between natural teeth and ceramics based on the input and output data of the multi-layer perceptron model; store the repaired tooth spectrum data of the cases and the corresponding model matching results in the optical database as a reference for doctors to query.

[0016] Preferably, it further includes the steps: S7. Based on the optical database, use the Flask framework and the ElementUI library to construct a web program for predicting the spectrum of repaired teeth and recommending ceramic matching; use the Flask framework to write the backend content of the web program, and use the ElementUI library to display and design the front-end page of the web program; the doctor submits the natural tooth spectrum data of the patient and the type of ceramic sample to be selected to the web program, and the web program feeds back the predicted repaired tooth spectrum data (for the doctor and patient to view and understand) and the matching ceramic sample type to the doctor.

[0017] Specifically, the data categories in all types of spectrum data include the average reflectance value and the CIE Lab color value.

[0018] Specifically, in step S1, after obtaining the natural tooth sample and performing pretreatment, use a hyperspectral imaging spectrometer to capture the hyperspectral reflectance image of the natural tooth sample; in the hyperspectral reflectance image, divide the natural tooth sample into multiple regions according to the preset regional positions, and respectively obtain the spectrum data of each region at different wavelengths, and then the natural tooth spectrum data can be obtained.

[0019] Specifically, in step S1, after obtaining the natural tooth spectrum data, use a turbine to separate the dentin and enamel of the natural tooth, and respectively make double-sided parallel thin slice samples, and complete fine polishing on a polishing machine to form the optical surfaces corresponding to the dentin and enamel; then use a hyperspectral imaging spectrometer to capture the reflectance images of the optical surfaces of the two to obtain the dentin spectrum data and the enamel spectrum data.

[0020] Further, in step S2, use the extinction coefficient to calibrate the semi-permeability of the enamel and ceramic samples. Specifically: measure the refractive index n, transmittance T and thickness t of the enamel and ceramic samples. First, establish the relationship between the refractive index n and the measured light wavelength λ using the Cauchy formula, as shown in the following formula:

[0021]

[0022] In the formula, B, C, and D represent dimensionless parameters; then the Lambert-Beer law is used to obtain the extinction coefficient corresponding to different measured light wavelengths λ, as shown in the following formula:

[0023]

[0024] In the formula, e represents the base of the natural logarithm, and μ represents the value of the extinction coefficient; thus, the characterization of the extinction coefficient for semi-permeability is realized.

[0025] Specifically, in step S4, the constructed multi-layer perceptron model includes two connected fully-connected layers, as well as an input layer and an output layer; the first fully-connected layer receives the extended input vector with the first dimension from the input layer and maps it to a vector with the second dimension; the second fully-connected layer receives the vector with the second dimension and maps it to a vector with the third dimension, and the output layer then maps the vector with the third dimension to output a result vector.

[0026] Preferably, the activation functions of the two fully-connected layers are both ReLU activation functions, and the loss functions are both mean squared errors; in step S2, the extinction coefficients of the enamel and ceramic samples are converted into normalized values to form a semi-permeability constraint vector; in step S5, it is concatenated with the initial input vector and then input into the multi-layer perceptron model; in the result vector, only the vectors that meet the threshold set for the corresponding dimension of the semi-permeability constraint vector are retained, corresponding to forming the model matching result.

[0027] The present invention also provides an optical prediction matching system for oral ceramic restorations based on deep learning, including:

[0028] A spectral data acquisition module, configured to acquire natural tooth spectral data, extract dentin spectral data and enamel spectral data respectively, and at the same time acquire ceramic spectral data of different types, as well as acquire the composite spectral data of the ceramic-dentin complex;

[0029] A semi-permeability calibration module, with an extinction coefficient calculation unit built in, used to calibrate the semi-permeability parameters respectively according to the enamel spectral data and the ceramic spectral data, and combine the semi-permeability parameters of the two to form a semi-permeability constraint vector;

[0030] A complex construction device, including a positioning fixture and an adhesion control unit, used to bond the ceramic sample corresponding to the ceramic spectral data to the separated dentin to obtain a ceramic-dentin complex;

[0031] A multi-layer perceptron construction module, used to construct a multi-layer perceptron model, and through the spectral data processing interface and the deep learning training engine, receive all types of spectral data, complete the training of the multi-layer perceptron model, and construct an initial input vector by means of feature splicing;

[0032] The intelligent matching decision module, including a vector fusion processor and a constraint condition screening unit, is used to splice the initial input vector and the semi-permeable constraint vector to form an extended input vector, and input the extended input vector into the trained multi-layer perceptron model to perform optical matching prediction; it is also used to perform threshold screening by combining the semi-permeability parameters of enamel and ceramic samples in the semi-permeable constraint vector during optical matching prediction, and output the model matching result that meets the preset constraint conditions.

[0033] In summary, due to the adoption of this technical solution, the beneficial effects of the present invention are as follows:

[0034] The present invention can effectively replace the traditional manual color matching and matching trial-and-error process that relies on doctors' experience, so as to achieve rapid matching of natural teeth and ceramic materials. After the natural tooth data is uploaded to the system of the present invention, it can be automatically matched with a variety of ceramic materials, with high matching efficiency.

[0035] The present invention makes full use of the advantages of deep learning technology, and can eliminate subjective factors of manual operations, such as visual fatigue and experience differences, through the multi-layer perceptron model, ensuring the consistency of matching results between different doctors or at different times by the same doctor, thereby reducing the rework rate. The output result of the model can also be quantified as a curve relationship diagram with wavelength as the abscissa and reflectivity as the ordinate, which is convenient for patients to objectively perceive the composite tooth parameters after restoration.

[0036] In addition, the present invention achieves a high-precision matching prediction effect. Through the non-linear modeling ability of the multi-layer perceptron model, it accurately learns the complex mapping relationship between natural teeth-ceramic materials and the restored composite teeth, and the model incorporates a semi-permeability index to further guide doctors in the selection of ceramics during the process of using ceramic materials to repair natural teeth. The current index of the multi-layer perceptron model on the MSE loss function is better than that of 1D CNN model, Transformer model, residual neural network model, etc. Its loss on the training set after iteration is 0.0098, and the loss on the test set is 0.032; when the absolute value of the difference between the predicted result and the actual result of reflectivity is defined as less than 0.001 as accurate, the accuracy of this model can reach 84%, and the experimental data shows that it has the best accuracy among the comparable models of the same kind. Brief Description of the Drawings

[0037] Figure 1 It is a schematic diagram of the overall process of the optical prediction matching method for oral ceramic restorations of the present invention;

[0038] Figure 2 It is a schematic diagram of the hyperspectral reflection image captured from natural tooth samples of the present invention;

[0039] Figure 3Schematic diagram of the division of natural tooth samples by regional location and their spectral data in the present invention;

[0040] Figure 4 Schematic diagram of the curve of the extinction coefficient corresponding to different measurement light wavelengths obtained by using the Lambert-Beer law in the present invention;

[0041] Figure 5 Schematic diagram of the structural connection relationship of the multi-layer perceptron model in the present invention;

[0042] Figure 6 Schematic diagram of the comparison between the predicted value and the true value of the optical reflectance of the composite restoration tooth by the system at different light wavelengths in the present invention. Detailed implementation manners

[0043] 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 below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Usually, the components of the embodiments of the present invention described and illustrated herein can be arranged and designed in various different configurations.

[0044] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but merely represents selected 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 fall within the scope of protection of the present invention.

[0045] Embodiment 1

[0046] An optical prediction matching method for oral ceramic restorations based on deep learning, Figure 1 shows a brief overview of the overall process of this method, which can be referred to synchronously; the key steps of this method are summarized as follows:

[0047] S1. Obtain the spectral data of natural teeth, and then separate the dentin and enamel of the natural teeth to obtain the spectral data of dentin and enamel respectively; obtain the spectral data of different types of ceramics;

[0048] S2. Use the extinction coefficient to calibrate the semi-permeability of enamel and the semi-permeability of the ceramic samples corresponding to the ceramic spectral data, and the calibration results of the two are combined to form a semi-permeability constraint vector;

[0049] S3. Bond the ceramic samples to the separated dentin to obtain a ceramic-dentin composite; obtain the composite spectral data of the corresponding ceramic-dentin composite;

[0050] S4. Construct a multi-layer perceptron model, and based on all the acquired spectral data of all types, conduct model training; then splice the dentin spectral data and the ceramic spectral data to form an initial input vector;

[0051] S5. Splice the initial input vector with the semi-permeability constraint vector to form an extended input vector, and input the extended input vector into the trained multi-layer perceptron model to obtain a model matching result.

[0052] Based on the above steps, in this embodiment, the composite spectral data of the ceramic-dentin composite and the natural tooth spectral data are optically predicted and matched by the trained multi-layer perceptron model, and the semi-permeability of the calibrated enamel and ceramic samples is used as a constraint condition to obtain the type of ceramic sample that meets the constraint condition, and the output is the model matching result.

[0053] In this embodiment, the data categories in all types of spectral data all include the average reflectance value and the CIE Lab color value.

[0054] As a preference of this embodiment, after step S5, it further includes step S6: Based on the input and output data of the multi-layer perceptron model, establish an optical database regarding the natural tooth-ceramic mapping relationship, store the corresponding data, including the spectral data of the restored tooth (this data is finally presented in CIE Lab color values for the convenience of users to view and understand); store the spectral data of the restored tooth and the corresponding model matching results of the cases in the optical database as a reference for doctors to query.

[0055] As a preference of this embodiment, after step S6, it further includes step S7: Based on the optical database, construct a web program for predicting the spectral data of the restored tooth and recommending ceramic matching through the Flask framework and the ElementUI library; use the Flask framework to write the backend content of the web program, and use the ElementUI library to display and design the front-end page of the web program;

[0056] The doctor submits the spectral data of the patient's natural tooth and the type of ceramic sample to be selected to the web program, and the web program feeds back the predicted spectral data of the restored tooth and the matching ceramic sample type to the doctor, so as to facilitate the doctor's selection and non-invasive restoration tooth design.

[0057] Embodiment 2

[0058] On the basis of Embodiment 1, this embodiment details the step details in the optical prediction and matching method for oral ceramic restorations and introduces preferred examples.

[0059] First, in step S1, healthy natural tooth samples extracted from patients due to reasons such as impacted teeth, orthodontics, and periodontitis can be prepared and pre-treated. In this embodiment, this batch of samples is rinsed with 0.9% isotonic saline, and after removing soft tissues such as gums and granulation, they are stored in a 0.1% thymol solution at 4°C.

[0060] Subsequently, using a hyperspectral imaging spectrometer, in this embodiment, a Specim IQ imaging spectrometer, hyperspectral reflectance images as Figure 2 shown are captured from the natural tooth samples. The image in this embodiment has a wavelength range from 400 nm to 1000 nm, and a spectral resolution of approximately 3 nm, with a total of 204 spectral bands, and a spatial resolution of 512×512 pixels. When taking pictures in this embodiment, in order to maximize the size of the area captured by the camera, the camera is placed at a position about 15 cm away from the natural tooth sample, and a CIE D65 daylight light source is used. The line integration time for each capture is set to 40 ms, and the image capture time is approximately 8 s to ensure good exposure without generating oversaturated pixel responses.

[0061] In this embodiment, in the hyperspectral reflectance image, Specim Insight software is used for data acquisition and processing. The natural tooth sample is divided into three regions as Figure 3 shown, namely: the cervical third, the middle third, and the incisal third. After obtaining the spectral data of each region at different wavelengths respectively, the natural tooth spectral data is obtained.

[0062] Also in step S1, after obtaining the natural tooth spectral data, a turbine is used to separate the dentin and enamel of the natural tooth, and double-sided parallel thin slice samples are respectively made, and fine polishing is completed on a polishing machine to form optical surfaces corresponding to the dentin and enamel; then a hyperspectral imaging spectrometer is used to capture the reflectance images of the optical surfaces of the two, and the dentin spectral data and the enamel spectral data are obtained.

[0063] The acquisition method of the spectral data of different types of ceramics also uses a hyperspectral imaging spectrometer, and is the same as the camera setting method for obtaining hyperspectral reflectance images in the natural tooth sample.

[0064] Next, in step S2, the extinction coefficient is used to calibrate the semi-permeability of the enamel and ceramic samples to avoid the influence of factors such as the environment and thickness. The specific method in this embodiment is as follows: an ellipsometer is used to measure the refractive index n of the enamel and ceramic samples, a spectrometer is used to measure the transmittance T of the enamel and ceramic samples, and a micrometer is used to measure the thickness t of the enamel and ceramic samples; the Cauchy formula is used to establish the relationship between the refractive index n and the measured light wavelength λ, as shown in the following formula:

[0065]

[0066] In the formula, B, C, and D represent dimensionless parameters used to describe the optical properties of enamel or ceramic samples; then the Lambert-Beer law is used to obtain the extinction coefficient corresponding to different measured light wavelengths λ, as shown in the following formula:

[0067]

[0068] In the formula, e represents the base of the natural logarithm, and μ represents the value of the extinction coefficient; thus, the characterization of the extinction coefficient for semi-permeability is realized, and reference can be made to Figure 4 for the schematic illustration. Figure 4 is the schematic curve obtained by using the Lambert-Beer law, where μ ext represents the total extinction coefficient, and μ bire represents the extinction coefficient caused by grain boundary birefringence, and μ pore represents the extinction coefficient caused by pores.

[0069] For step S3, in this embodiment, a commonly used dental resin adhesive is coated, and the corresponding ceramic sample is bonded to the dentin. After cleaning the excess adhesive at the edge, light curing is carried out for 30 s to obtain a ceramic-dentin composite. The hyperspectral imaging spectrometer is also used to obtain its composite spectral data.

[0070] Since step S4 requires the training of a multi-layer perceptron model (MLP) and the matching task requirements for different situations such as natural teeth and ceramic material types, taking dentin under the same batch category as an example, in this embodiment, by replacing different types of ceramic materials, multiple ceramic-dentin composites and their composite spectral data can be obtained; then by replacing another batch of natural teeth and performing the same processing operations, the sample data can be expanded to meet the needs of model training.

[0071] In step S4, as Figure 5 shown, the constructed multi-layer perceptron model includes two connected fully connected layers ( Figure 5 the hidden layer part in), as well as an input layer and an output layer. In this embodiment, after the dentin spectral data and the ceramic spectral data are spliced, an initial input vector (408, 2) is formed, which is further spliced with the semi-permeability constraint vector to form an extended input vector and input into the multi-layer perceptron model.

[0072] In the multi-layer perceptron model, the first fully connected layer receives the 408-dimensional extended input vector from the input layer and maps it to a 512-dimensional second-dimensional vector, where the ReLU activation function is used here; the second fully connected layer maps this 512-dimensional second-dimensional vector to a 256-dimensional third-dimensional vector, and the ReLU activation function is also used here; finally, the output layer maps this 256-dimensional third-dimensional vector to a 204-dimensional result vector.

[0073] In this embodiment, the loss functions of the two fully-connected layers are both mean squared error (MSELoss), which is specifically designed for regression problems and can quantify the gap between the predicted value and the actual value. Further in this embodiment, in step S2, the extinction coefficients of the enamel and ceramic samples can be converted into normalized values (such as in the range of 0 to 1) and used as an additional conditional vector, that is, a semi-permeability constraint vector. Then, in step S5, it is concatenated with the initial input vector and input into the multi-layer perceptron model.

[0074] For the result vector, this embodiment only retains the vectors that meet the threshold set for the corresponding dimension of the semi-permeability constraint vector (such as setting a constraint condition that the semi-permeability error is less than 5%), and correspondingly forms the model matching result, thus obtaining the type of ceramic sample that meets the constraint conditions.

[0075] Embodiment 3

[0076] Based on any of the above embodiments, this embodiment introduces the system basis for the application of the method, that is, an optical prediction matching system for oral ceramic restorations based on deep learning. The system includes:

[0077] A spectral data acquisition module, configured to obtain the spectral data of natural teeth, extract the spectral data of dentin and enamel respectively, and at the same time obtain the spectral data of different types of ceramics, as well as obtain the composite spectral data of ceramic-dentin composites;

[0078] A semi-permeability calibration module, which has an extinction coefficient calculation unit built in, and is used to calibrate the semi-permeability parameters according to the enamel spectral data and the ceramic spectral data respectively, and combine the semi-permeability parameters of the two to form a semi-permeability constraint vector;

[0079] A composite body construction device, which includes a positioning fixture and an adhesion control unit, and is used to bond the ceramic sample corresponding to the ceramic spectral data to the separated dentin to obtain a ceramic-dentin composite;

[0080] A multi-layer perceptron construction module, which is used to construct a multi-layer perceptron model, and through the spectral data processing interface and the deep learning training engine, receive all types of spectral data, complete the training of the multi-layer perceptron model, and construct an initial input vector by means of feature splicing;

[0081] An intelligent matching decision module, which includes a vector fusion processor and a constraint condition screening unit, and is used to concatenate the initial input vector and the semi-permeability constraint vector to form an extended input vector, and input the extended input vector into the trained multi-layer perceptron model to perform optical matching prediction; it is also used to perform threshold screening by combining the semi-permeability parameters of the enamel and the ceramic sample in the semi-permeability constraint vector during optical matching prediction, and output the model matching result that meets the preset constraint conditions.

[0082] Figure 6 It shows the comparison between the predicted value and the true value of the optical reflectance of the composite restored tooth by the system at different light wavelengths after applying the optical prediction matching method for oral ceramic restorations; among which the blue curve represents the true value and the red curve represents the predicted value. Only the reflectance values in the visible light wavelength range need to be concerned, that is, the wavelength range on the abscissa is the interval from 400nm to 760nm. The curve coincidence degree in this interval is relatively high, the loss on the training set is 0.0098 after iteration, the loss on the test set is 0.032, and the accuracy of the system model can reach 84%, representing a relatively high precision level.

Claims

1. An optical prediction matching method for oral ceramic restorations based on deep learning, characterized in that, It includes the following steps: S1. Obtain the spectral data of natural teeth. Subsequently, separate the dentin and enamel of the natural teeth, and obtain the dentin spectral data and enamel spectral data respectively; obtain the spectral data of different types of ceramics; S2. Use the extinction coefficient to calibrate the semi-permeability of the enamel and the semi-permeability of the ceramic samples corresponding to the ceramic spectral data. The calibration results of the two are combined to form a semi-permeability constraint vector; S3. Bond the ceramic samples to the separated dentin to obtain a ceramic-dentin composite; obtain the composite spectral data corresponding to the ceramic-dentin composite; S4. Construct a multi-layer perceptron model, and perform model training based on all the obtained spectral data of different types; then splice the dentin spectral data and the ceramic spectral data to form an initial input vector; S5. Splice the initial input vector and the semi-permeability constraint vector to form an extended input vector, and input the extended input vector into the trained multi-layer perceptron model to obtain a model matching result; In step S5, the trained multi-layer perceptron model performs optical prediction matching on the composite spectral data of the ceramic-dentin composite and the spectral data of the natural teeth, and uses the calibrated semi-permeability of the enamel and the ceramic samples as constraint conditions to obtain the type of ceramic samples that meet the constraint conditions, and the output is the model matching result.

2. The optical prediction matching method for oral ceramic restorations according to claim 1, wherein, It also includes the step: S6. Establish an optical database on the mapping relationship between natural teeth and ceramics based on the input and output data of the multi-layer perceptron model; Store the spectral data of the restored teeth and the corresponding model matching results of the cases in the optical database as a reference for doctors to query.

3. The optical prediction matching method for oral ceramic restorations according to claim 2, wherein It also includes the step: S7. Based on the optical database, use the Flask framework and the ElementUI library to construct a web program for predicting the spectral data of restored teeth and recommending ceramic matching; use the Flask framework to write the backend content of the web program, and use the ElementUI library to display and design the front-end page of the web program; the doctor submits the spectral data of the patient's natural teeth and the type of ceramic samples to be selected to the web program, and the web program feeds back the predicted spectral data of the restored teeth and the matching type of ceramic samples to the doctor.

4. The method for optically predicting and matching an oral ceramic restoration according to claim 1, characterized in that: The data categories in all types of spectral data include the average reflectance value and the CIE Lab color value.

5. The optical prediction matching method for oral ceramic restorations according to claim 1, wherein: In step S1, after obtaining the natural tooth sample and performing pretreatment, use a hyperspectral imaging spectrometer to capture the hyperspectral reflectance image of the natural tooth sample; in the hyperspectral reflectance image, divide the natural tooth sample into multiple regions according to the preset regional positions, and obtain the spectral data of each region at different wavelengths, and then the spectral data of the natural teeth can be obtained.

6. The optical prediction matching method for oral ceramic restorations according to claim 5, characterized in that: In step S1, after obtaining the spectral data of the natural teeth, use a turbine to separate the dentin and enamel of the natural teeth, and make them into double-sided parallel thin sheet samples respectively, and complete fine polishing on a grinding and polishing machine to form the optical surfaces corresponding to the dentin and enamel; then use a hyperspectral imaging spectrometer to capture the reflectance images of the optical surfaces of the two to obtain the dentin spectral data and the enamel spectral data.

7. The optical prediction matching method for oral ceramic restorations according to claim 1, characterized in that In step S2, the semi - transparency of the enamel and ceramic samples is calibrated using the extinction coefficient. Specifically: the refractive index n, transmittance T, and thickness t of the enamel and ceramic samples are measured. First, the relationship between the refractive index n and the measurement light wavelength λ is established using the Cauchy formula as follows: In the formula, B, C, and D represent dimensionless parameters; then, the Lambert - Beer law is used to obtain the corresponding extinction coefficients at different measurement light wavelengths λ as follows: In the formula, e represents the base of the natural logarithm, and μ represents the value of the extinction coefficient; thus, the characterization of the semi - transparency by the extinction coefficient is realized.

8. The optical prediction matching method for oral ceramic restorations according to claim 1, characterized in that: In step S4, the constructed multi - layer perceptron model includes two connected fully - connected layers, as well as an input layer and an output layer; the first fully - connected layer receives the extended input vector with the first dimension from the input layer and maps it to a vector with the second dimension; the second fully - connected layer receives the vector with the second dimension and maps it to a vector with the third dimension, and the output layer then maps the vector with the third dimension to output a result vector.

9. The method for optically predicting and matching an oral ceramic restoration according to claim 8, wherein: The activation functions of the two fully - connected layers are both ReLU activation functions, and the loss functions are both mean square errors; in step S2, the extinction coefficients of the enamel and ceramic samples are converted into normalized values to form a semi - transparency constraint vector; in step S5, it is concatenated with the initial input vector and then input into the multi - layer perceptron model; in the result vector, only the vectors that meet the threshold set for the corresponding dimension of the semi - transparency constraint vector are retained, corresponding to forming the model matching result.

10. An optical prediction matching system for oral ceramic restorations based on deep learning, characterized in that, Including: A spectral data acquisition module, configured to obtain the spectral data of natural teeth, extract the spectral data of dentin and enamel respectively, and at the same time obtain the spectral data of different types of ceramics, as well as obtain the composite spectral data of the ceramic - dentin complex; A semi - transparency calibration module, with an extinction coefficient calculation unit built - in, used to calibrate the semi - transparency parameters respectively according to the enamel spectral data and the ceramic spectral data, and combine the semi - transparency parameters of the two to form a semi - transparency constraint vector; A complex construction device, including a positioning fixture and an adhesion control unit, used to bond the ceramic sample corresponding to the ceramic spectral data onto the separated dentin to obtain a ceramic - dentin complex; A multi - layer perceptron construction module, used to construct a multi - layer perceptron model, and through the spectral data processing interface and the deep learning training engine, receive all types of spectral data, complete the training of the multi - layer perceptron model, and construct an initial input vector by means of feature concatenation; An intelligent matching decision module, including a vector fusion processor and a constraint condition screening unit, used to concatenate the initial input vector and the semi - transparency constraint vector to form an extended input vector, and input the extended input vector into the trained multi - layer perceptron model to perform optical matching prediction; It is also used to perform threshold screening by combining the semi - transparency parameters of the enamel and ceramic samples in the semi - transparency constraint vector during optical matching prediction, and output a model matching result that meets the preset constraint conditions.

Images