Plaque quantification system, method and storage medium

Abstract

The application relates to the technical field of oral cavity equipment, and discloses a dental plaque quantitative analysis system, a method and a storage medium. The system comprises a dental plaque observation instrument and an upper computer. The dental plaque observation instrument is used for collecting a two-dimensional fluorescence image of teeth in an oral cavity area of a to-be-measured object. The upper computer is in communication connection with the dental plaque observation instrument. The upper computer is configured to: receive the two-dimensional fluorescence image of teeth sent by the dental plaque observation instrument; perform image segmentation and image depth estimation based on the two-dimensional fluorescence image of teeth, perform three-dimensional reconstruction, and obtain a three-dimensional tooth model containing a plaque area and a tooth area; then calculate the surface area of the plaque area and the surface area of the tooth area respectively, obtain an area ratio of the surface area of the plaque area to the surface area of the tooth area, and take the area ratio as a quantitative analysis result of dental plaque. The application can accurately calculate the area of the plaque area and realize high-precision quantitative detection of dental plaque.

Description

Technical Field

[0001] This application relates to the field of dental equipment technology, and in particular to a system, method and storage medium for quantitative analysis of dental plaque. Background Technology

[0002] Dental plaque is a major risk factor for various oral diseases such as tooth decay and periodontitis. Its formation process is insidious and its area is easily spread; therefore, accurate identification and quantitative analysis of dental plaque are crucial for the prevention and treatment of oral diseases. Currently, dental plaque developers combined with optical observation instruments are widely used in clinical and home settings to observe the tooth surface, assessing plaque condition by the size and distribution of the stained areas. However, existing dental plaque detection methods still rely heavily on the visual judgment of dentists or users, which is highly subjective and difficult to quantify, hindering long-term monitoring of plaque trends.

[0003] To improve the objectivity of detection, existing technologies have proposed using two-dimensional fluorescence imaging to automatically identify plaque areas. However, on the one hand, two-dimensional images can only reflect the projected information of the tooth surface and cannot accurately describe the complex true morphology of the tooth surface. For example, structures such as interdental spaces, grooves, and local depressions will produce obvious geometric distortions in two-dimensional images, causing plaque area to be unable to be accurately calculated from pixel area to the true surface area, thus leading to significant quantitative bias. On the other hand, if the fluorescence signal collected by the dental plaque observation instrument is interfered with by unfiltered light from the excitation source, it will appear as abnormal brightness in the two-dimensional fluorescence image, masking the true fluorescence image features and further affecting the subsequent quantitative analysis results.

[0004] Therefore, how to conduct accurate and reliable quantitative analysis of dental plaque has become a pressing technical problem in the field of dentistry. Summary of the Invention

[0005] Therefore, it is necessary to address the problem of low accuracy in existing quantitative detection of dental plaque by proposing a quantitative analysis system, method, and storage medium for dental plaque.

[0006] A first aspect of this application provides a dental plaque quantitative analysis system, the system comprising a dental plaque observer and a host computer, the dental plaque observer being used to acquire two-dimensional fluorescence images of teeth in the oral cavity region of a subject, the host computer being communicatively connected to the dental plaque observer, and the host computer being configured to:

[0007] Receive two-dimensional fluorescence images of teeth transmitted by the dental plaque observation instrument;

[0008] Based on the two-dimensional fluorescence image of the teeth, the plaque region and the tooth region are segmented to obtain the plaque segmentation result and the tooth segmentation result.

[0009] Two-dimensional image depth estimation is performed on the two-dimensional fluorescence image of the tooth to obtain a relative depth map that characterizes the spatial information of the tooth surface.

[0010] Based on the plaque segmentation results, tooth segmentation results, and relative depth map, a three-dimensional reconstruction is performed to obtain a three-dimensional tooth model containing plaque and tooth regions.

[0011] The surface of the three-dimensional tooth model is triangulated, and the surface area of ​​the plaque region is calculated based on the triangulation results. The surface area of ​​the tooth region is also calculated, and the area ratio of the surface area of ​​the plaque region to that of the tooth region is obtained. The area ratio is used as the quantitative analysis result of dental plaque.

[0012] Furthermore, the dental plaque observation instrument includes an observation instrument body, an optical lens group, a light source, an image sensor, a transparent window, a filter, and an annular spacer. The light source is configured to switch between ultraviolet light and white light.

[0013] The main body of the observation instrument has an opening, the optical lens group is disposed within the opening, the light source is disposed around the light-incident end of the optical lens group, the image sensor is disposed at the rear end of the optical lens group, a transparent window is disposed at the opening of the opening, the transparent window is sealed to the main body of the observation instrument, the transparent window covers at least the light-incident end of the optical lens group and the light-emitting surface of the light source, and a filter is disposed between the transparent window and the image sensor;

[0014] An annular spacer is provided between the transparent window and the optical lens assembly. One side of the annular spacer abuts against the surface of the transparent window and the other side of the annular spacer abuts against the surface of the optical lens assembly. The annular spacer is made of an opaque material.

[0015] Furthermore, at least one side of the transparent window sheet is coated with a light-shielding layer at a position opposite to the annular spacer, and the light-shielding layer is annular.

[0016] Furthermore, the projection of the annular spacer onto the transparent window sheet does not extend beyond the area where the light-shielding layer is located.

[0017] Further, the step of the host computer performing image segmentation of the plaque region and the tooth region based on the two-dimensional fluorescence image of the tooth to obtain plaque segmentation results and tooth segmentation results includes:

[0018] The host computer calls a pre-trained dental plaque segmentation model to perform image segmentation on the two-dimensional fluorescence image of the teeth, and outputs plaque segmentation results and tooth segmentation results; wherein, the dental plaque segmentation model is obtained through the following training method:

[0019] A training dataset and a validation dataset are constructed. Each training dataset contains several training images. Each training image is a two-dimensional fluorescence image of teeth, and each training image has been pre-segmented and labeled with the tooth region and the dental plaque region.

[0020] An initial segmentation model is constructed, which adopts the TransUNet architecture;

[0021] The training images in the training dataset are input into the initial segmentation model to obtain the model's prediction results for dental plaque and tooth regions.

[0022] The loss value between the model's prediction results and the labeled results is calculated using the following loss function: ,in, and These are the weighting coefficients of the loss function. , , The number of dental plaque pixels correctly identified by the model. The number of non-dental plaque pixels that the model incorrectly identified as dental plaque. This represents the number of dental plaque pixels not identified by the model. This is the difficulty weighting coefficient. , This is the penalty coefficient;

[0023] The parameters of the initial segmentation model are updated iteratively using the backpropagation algorithm to minimize the loss value, thereby obtaining a transient segmentation model.

[0024] During training, the transient segmentation model is used to test and verify the validation dataset. If the validation result meets the preset iteration cutoff condition, the validated transient segmentation model is used as the trained dental plaque segmentation model.

[0025] Furthermore, the system also includes a display, which is communicatively connected to the host computer. The host computer performs a step of three-dimensional reconstruction based on the plaque segmentation results, tooth segmentation results, and the relative depth map to obtain a three-dimensional tooth model containing plaque regions and tooth regions, including:

[0026] The host computer constructs a point cloud map of the tooth surface based on the relative depth map;

[0027] Based on the plaque segmentation results and the tooth segmentation results, different textures are applied to the corresponding regions in the point cloud image to generate a three-dimensional tooth model containing plaque regions and tooth regions.

[0028] The three-dimensional model of the teeth is displayed on the monitor.

[0029] Further, the host computer performs triangulation on the surface of the three-dimensional tooth model, calculates the surface area of ​​the plaque region based on the triangulation results, and calculates the surface area of ​​the tooth region, including:

[0030] The host computer extracts the corresponding surface point clouds from the three-dimensional tooth model based on the plaque segmentation results and the tooth segmentation results, respectively, to obtain a plaque cloud set and a tooth point cloud set.

[0031] Delaunay triangulation is performed on the bacterial spot cloud set and the tooth point cloud set respectively to form a bacterial spot region triangular mesh and a tooth region triangular mesh respectively, wherein the bacterial spot region triangular mesh and the tooth region triangular mesh each contain a number of triangular pieces;

[0032] Based on the plaque segmentation results and the tooth segmentation results, the boundary point set is determined. The triangular pieces in the triangular mesh of the plaque region and the triangular mesh of the tooth region are screened respectively. The triangular pieces whose three vertices all belong to the boundary point set are regarded as hollow triangular pieces and the hollow triangular pieces are removed.

[0033] The surface area of ​​the plaque region is obtained by summing the areas of the triangular pieces in the plaque region after removing the cavities; and the surface area of ​​the tooth region is obtained by summing the areas of the triangular pieces in the tooth region after removing the cavities.

[0034] Further, the host computer performs Delaunay triangulation on the plaque cloud set and the tooth point cloud set respectively, forming plaque region triangular meshes and tooth region triangular meshes respectively, including:

[0035] The host computer takes any target point in the bacterial spot cloud set and the tooth point cloud set as the center, and selects a preset number of neighboring points around the target point using the K-nearest neighbor algorithm to form the neighborhood point set of the center.

[0036] The coordinates of the neighborhood point set are subtracted from the coordinates of the target point for centering, and a covariance matrix is ​​constructed.

[0037] The eigenvector corresponding to the smallest eigenvalue of the covariance matrix is ​​obtained by principal component analysis and used as the normal vector of the target point.

[0038] For any two target points to be connected, obtain their corresponding normal vectors and calculate the spatial angle between the two normal vectors.

[0039] Determine whether the spatial angle is less than or equal to a preset angle threshold;

[0040] If so, connect the two target points to be connected to form the effective edges of the triangulation;

[0041] Based on all valid edges connected in the set of bacterial spot clouds, a triangular mesh for the bacterial spot region is formed; and based on all valid edges connected in the set of tooth point clouds, a triangular mesh for the tooth region is formed.

[0042] A second aspect of this application provides a method for quantitative analysis of dental plaque, the method comprising:

[0043] Receive two-dimensional fluorescent images of teeth sent by a dental plaque detector;

[0044] Based on the two-dimensional fluorescence image of the teeth, the plaque region and the tooth region are segmented to obtain the plaque segmentation result and the tooth segmentation result.

[0045] Two-dimensional image depth estimation is performed on the two-dimensional fluorescence image of the tooth to obtain a relative depth map that characterizes the spatial information of the tooth surface.

[0046] Based on the plaque segmentation results, tooth segmentation results, and relative depth map, a three-dimensional reconstruction is performed to obtain a three-dimensional tooth model containing plaque and tooth regions.

[0047] The surface of the three-dimensional tooth model is triangulated, and the surface area of ​​the plaque region is calculated based on the triangulation results. The surface area of ​​the tooth region is also calculated, and the area ratio of the surface area of ​​the plaque region to that of the tooth region is obtained. The area ratio is used as the quantitative analysis result of dental plaque.

[0048] A third aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to perform the steps of the above-described method for quantitative analysis of dental plaque.

[0049] This application discloses a dental plaque quantitative analysis system, method, and storage medium, which realizes the quantitative analysis of dental plaque from two-dimensional fluorescence images to three-dimensional surface area. The system acquires two-dimensional fluorescence images and automatically identifies plaque areas. It constructs a three-dimensional model of the tooth surface through monocular depth estimation and accurately calculates the surface area of ​​the plaque area and the tooth area in three-dimensional space, effectively avoiding area distortion caused by two-dimensional projection. This dental plaque quantitative analysis system can reflect the accurate location and true surface morphology of teeth, improve the accuracy of plaque area calculation, and achieve high-precision quantitative detection of dental plaque. This allows for an objective reflection of the relationship between dental plaque and disease, which is of great significance for disease prevention, diagnosis, treatment, and prognosis. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] in:

[0052] Figure 1 This is a schematic diagram of the structure of a dental plaque quantitative analysis system in one embodiment;

[0053] Figure 2 This is a flowchart illustrating a method for quantitative analysis of dental plaque in one embodiment;

[0054] Figure 3 This is a schematic diagram of the axial structure of a dental plaque observation device in one embodiment;

[0055] Figure 4 This is a top view of the dental plaque observation device in one embodiment;

[0056] Figure 5 for Figure 4 A schematic diagram of the cross-sectional structure corresponding to region A of the dental plaque observation instrument in the image;

[0057] Figure 6 This is a schematic diagram of the planar structure of a transparent window in one embodiment;

[0058] Figure 7 Comparison of two-dimensional fluorescence images of teeth acquired by a dental plaque observation instrument with and without annular septa. Figure 7 (a) A two-dimensional fluorescence image of teeth acquired using a dental plaque detector without an annular septum. Figure 7 (b) A two-dimensional fluorescent image of teeth acquired by a dental plaque observation instrument with an annular septum in one embodiment;

[0059] Figure 8 Here is a two-dimensional fluorescence image of a tooth and its corresponding segmentation image in one embodiment, wherein, Figure 8 (a) is a two-dimensional fluorescence image of the teeth. Figure 8 (b) is the corresponding segmented image;

[0060] Figure 9 Here is a three-dimensional model image of teeth in one embodiment, wherein, Figure 9 (a) Figure 9 (b) and Figure 9 (c) Showing the 3D model from different angles;

[0061] Figure 10This is a schematic diagram of a plaque area in the concave surface of the gap between adjacent teeth in one embodiment.

[0062] Figure 11 for Figure 10 A schematic diagram of the distortion results of triangulation of the plaque area in the concave surface of the interdental space;

[0063] Figure 12 for Figure 10 A schematic diagram showing the correct triangulation results of the plaque area in the concave surface of the tooth gap;

[0064] Figure 13 This is a structural block diagram of a computer device in one embodiment.

[0065] Explanation of the attached drawing numbers:

[0066] 100: Dental plaque observer; 200: Host computer; 110: Observation instrument body; 120: Optical lens group; 130: Light source; 140: Image sensor; 150: Transparent window; 160: Filter; 170: Annular spacer; 180: Light-shielding layer. Detailed Implementation

[0067] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0068] Figure 1 This is a schematic diagram of a dental plaque quantitative analysis system in one embodiment. The system includes a dental plaque observer 100 and a host computer 200. The dental plaque observer 100 and the host computer 200 are connected via wired or wireless means. The dental plaque observer 100 is used to acquire two-dimensional fluorescence images of teeth in the oral cavity of the subject, and the host computer 200 is used to execute the steps of the dental plaque quantitative analysis method.

[0069] In this embodiment, the dental plaque observation instrument 100 can be handheld by a user or used by medical personnel during clinical operations. It acquires a two-dimensional fluorescent image of the tooth surface by irradiating it with excitation light. This image includes fluorescent images of both the tooth area and the dental plaque area, which are displayed in different colors. The acquired two-dimensional fluorescent images of the teeth are transmitted to a host computer 200 in real time. The host computer 200 automatically completes subsequent image processing, three-dimensional reconstruction, and area calculation operations to obtain the final quantitative analysis results of dental plaque.

[0070] The host computer is configured as follows: Figure 2 The process shown is as follows:

[0071] S1: Receive a two-dimensional fluorescent image of teeth sent by the dental plaque observation instrument;

[0072] S2: Based on the two-dimensional fluorescence image of the teeth, perform image segmentation on the plaque region and the tooth region to obtain plaque segmentation results and tooth segmentation results;

[0073] S3: Perform two-dimensional image depth estimation on the two-dimensional fluorescence image of the tooth to obtain a relative depth map used to characterize the spatial information of the tooth surface;

[0074] S4: Perform three-dimensional reconstruction based on the plaque segmentation results, tooth segmentation results, and relative depth map to obtain a three-dimensional tooth model containing plaque and tooth regions;

[0075] S5: Triangulate the surface of the three-dimensional tooth model, calculate the surface area of ​​the plaque region based on the triangulation results, calculate the surface area of ​​the tooth region, and obtain the area ratio of the surface area of ​​the plaque region to the surface area of ​​the tooth region. Use the area ratio as the quantitative analysis result of dental plaque.

[0076] In step S1 above, the host computer and the dental plaque observer can communicate via USB, Wi-Fi, Bluetooth, or other short-range wireless communication methods. During the detection process, the dental plaque observer illuminates the patient's oral cavity. Its light source can switch between white light and ultraviolet light modes. Ultraviolet light can excite the metabolic products in dental plaque to produce fluorescence, causing a significant brightness difference between the plaque area and the tooth surface. The optical lens group at the front end of the dental plaque observer is responsible for imaging the oral cavity scene. The image sensor captures real-time fluorescence images and converts them into digital image data. The two-dimensional fluorescence images of the teeth acquired by the dental plaque observer are sent to the host computer via the communication module. The host computer preprocesses the images, then performs buffering and subsequent processing.

[0077] The preprocessing steps for the two-dimensional fluorescence image of teeth include distortion correction of the two-dimensional fluorescence image of teeth, specifically including:

[0078] Multiple images were pre-acquired using a checkerboard or dot matrix approach. The intrinsic parameter matrix K and distortion parameters of the optical lens assembly for the dental plaque observation instrument were calculated using OpenCV's calibrateCamera() function. The distortion parameters included radial distortion parameters. , and and tangential distortion parameters and ;

[0079] For each pixel coordinate in a two-dimensional fluorescence image of a tooth First, convert it into normalized camera coordinates using the intrinsic parameter matrix K. The conversion formula is: , ,in , Focal length , Using the principal point coordinates, and then performing inverse distortion calculations on the normalized camera coordinates based on the distortion parameters, the distorted coordinates are obtained. The inverse distortion calculation includes radial distortion. , , Tangential distortion: , ;

[0080] The distorted coordinates ( The sampling coordinates are obtained by projecting the intrinsic parameter matrix K back to the original image. ,in, , This completes the distortion correction of the fluorescence image.

[0081] In step S2 above, the host computer performs automatic segmentation on the two-dimensional fluorescence image of the teeth obtained in step S1 to distinguish between plaque regions and normal tooth regions. In some embodiments, the host computer calls a pre-trained dental plaque segmentation model to perform pixel-level classification on the input image and outputs region labeling results with the same dimensions as the original image. The dental plaque segmentation model can adopt a deep learning architecture, such as a convolutional neural network. The model's input is the two-dimensional fluorescence image of the teeth, and the output is a category label for each pixel, used to indicate whether the pixel belongs to a plaque region or a tooth region. Finally, the host computer obtains the plaque region segmentation results and the tooth region segmentation results, respectively.

[0082] In step S3 above, the host computer performs depth estimation on the two-dimensional fluorescence image to recover the spatial geometry of the tooth surface. The depth estimation model can be based on monocular image depth prediction technology, which infers the relative depth value of each pixel by analyzing texture features, edge information, brightness changes, etc. in the image. A relative depth map with the same size as the original image is generated through depth estimation, and the depth value of each pixel represents its relative distance relationship in the three-dimensional scene.

[0083] In step S4 above, the host computer performs 3D reconstruction based on the plaque segmentation results, tooth segmentation results, and relative depth map. Specifically, the depth map projects each pixel in the 2D image into 3D space, constructing corresponding 3D point cloud coordinates based on the depth values. This point cloud can represent the overall geometric structure of the tooth surface, including complex morphologies such as interdental spaces, grooves, and local depressions. Then, based on the segmentation results, each point in the point cloud is assigned a regional attribute, such as a plaque or a tooth point. Finally, the original image is mapped as a texture map onto the 3D point cloud or 3D mesh surface, ensuring that the 3D model maintains visually consistent color characteristics with the 2D image, facilitating plaque distribution identification. The 3D tooth model can also provide a three-dimensional visualization of the tooth surface, facilitating observation of the actual plaque coverage on the 3D curved surface.

[0084] In step S5 above, the host computer calculates the surface area of ​​the constructed 3D model. In some embodiments, the 3D surface is divided into several small triangular pieces through triangulation, and the area of ​​each triangular piece is calculated based on the plaque area and the tooth area. The area of ​​the triangle can be calculated using the 3D vector cross product method. After summing all the triangular piece surfaces, the surface area of ​​the plaque area and the surface area of ​​the tooth area can be obtained respectively. In other embodiments, the surface area of ​​the plaque area is calculated based on the triangulation results, while the surface area of ​​the tooth area is calculated using other surface area calculation methods. For example, it can be calculated using the total tooth area from the initial data collected in the system before the test object, and the total tooth area minus the surface area of ​​the plaque area measured this time is used as the surface area of ​​the tooth area measured this time. The ratio of the plaque surface area to the total tooth surface area is calculated as the quantitative analysis result of dental plaque, which can accurately reflect the degree of plaque coverage.

[0085] The dental plaque quantitative analysis system provided in this application realizes the quantitative analysis of dental plaque from two-dimensional fluorescence images to three-dimensional surface area. The system acquires two-dimensional fluorescence images and automatically identifies plaque areas. It constructs a three-dimensional model of the tooth surface through monocular depth estimation and accurately calculates the surface area of ​​the plaque area and the tooth area in three-dimensional space, effectively avoiding area distortion caused by two-dimensional projection. This application can reflect the accurate location and true surface morphology of teeth, improve the accuracy of plaque area calculation, and achieve high-precision quantitative detection of dental plaque. Furthermore, it can objectively reflect the relationship between dental plaque and disease, which is of great significance for the prevention, diagnosis, treatment, and prognosis of diseases.

[0086] In some specific embodiments, reference is made to Figures 3-5 The dental plaque observation instrument 100 includes an observation instrument body 110, an optical lens group 120, a light source 130, an image sensor 140, a transparent window 150, a filter 160, and an annular spacer 170. The light source 130 is configured to switch between ultraviolet light and white light.

[0087] The main body 110 of the observation instrument has an opening, the optical lens group 120 is disposed in the opening, the light source 130 is disposed around the light-incident end of the optical lens group 120, the image sensor 140 is disposed at the rear end of the optical lens group 120, a transparent window 150 is disposed at the opening of the opening, the transparent window 150 is sealed to the main body 110 of the observation instrument, the transparent window 150 at least covers the light-incident end of the optical lens group 120 and the light-emitting surface of the light source 130, and a filter 160 is disposed between the transparent window 150 and the image sensor 140;

[0088] An annular spacer 170 is provided between the transparent window 150 and the optical lens group 120. One side of the annular spacer 170 abuts against the surface of the transparent window 150, and the other side of the annular spacer 170 abuts against the surface of the optical lens group 120. The annular spacer 170 is made of an opaque material.

[0089] In this embodiment, the white light mode of the light source 130 is used for conventional visual observation, while the violet light mode is used to excite the metabolites in dental plaque, causing them to generate fluorescence signals. This facilitates the presentation of the differences between dental plaque and teeth in a two-dimensional fluorescence image of the teeth. The violet light wavelength is approximately 405 nm. Due to the limited space at the head of the dental plaque observer 100, the light source 130 is arranged in a ring around the light-incident end of the optical lens group 120 to ensure uniform illumination of the tooth surface and improve imaging brightness. However, this also results in a very close distance between the light source 130 and the optical lens group 120. The excitation light emitted by the light source 130 is easily reflected at a large angle by the transparent window 150 into the optical lens group 120. Since the excitation light energy is significantly higher than the fluorescence, a weak fluorescence signal may be masked by the excitation light signal. At the same time, the stray light generated by the reflection of the excitation light reduces the contrast of the fluorescence image, decreasing the color contrast between the teeth and plaque. Furthermore, the color of plaque or teeth in the two-dimensional fluorescence image of the teeth is no longer the expected orange-red and green, affecting the accuracy of image segmentation.

[0090] The image sensor 140 employs a high-sensitivity color CMOS or CCD.

[0091] The transparent window 150 is set at the opening of the main body and connected in a sealed manner to provide waterproof, anti-fouling and antibacterial protection, which can meet the IP68 protection level. In addition, the light-inlet end corresponding to the optical lens group 120 and the light-outlet surface of the light source 130 also need to maintain a high transparency state to avoid affecting the propagation of the light path.

[0092] Reflective filters (160°) are generally effective at blocking directly incident excitation light. However, for excitation light incident at large angles, the filter's effectiveness is greatly reduced, allowing this excitation light to still enter the lens and create abnormal brightness, severely affecting the imaging of fluorescence signals. Since fluorescence energy is usually weak, while excitation light is strong, these abnormal highlights can cover or mask the weak fluorescence in the plaque area, resulting in unclear plaque imaging and reduced contrast in the two-dimensional fluorescence image. This reduces the color contrast between the two and affects the accuracy of subsequent image analysis.

[0093] In an optional embodiment, an annular spacer 170 and a filter 160 are provided between the transparent window 150 and the optical lens group 120, with the filter 160 disposed within the annular spacer 170.

[0094] An annular spacer 170 is located between the transparent window 150 and the optical lens group 120, surrounding the edge of the light-transmitting aperture of the optical lens group 120. The annular spacer 170 is made of an opaque material; in some embodiments, it is made of EVA cotton, which is elastic and compressible, allowing it to better conform to the surfaces of the transparent window 150 and the optical lens group 120, reducing gaps caused by structural tolerances, and effectively preventing light from the light source 130 from leaking into the area of ​​the optical lens group 120 through the gaps. The annular spacer 170 can also effectively block excitation light incident at large angles (i.e., excitation light reflected at large angles by the transparent window 150), which is difficult to filter through the filter 160. Figure 7 (a) is a two-dimensional fluorescent image of teeth acquired by a dental plaque observer without an annular septum. The contrast between teeth and plaque is reduced, the color contrast is decreased, and the edges are difficult to distinguish. Figure 7 (b) A two-dimensional fluorescent image of teeth acquired by a dental plaque observation instrument equipped with an annular septum 170 according to an embodiment of this application. The teeth and plaque areas appear as expected in green and orange-red, respectively, and the image has high clarity, rich detail, high contrast between teeth and plaque, increased color contrast, and clearly distinguishable edges. Figure 7 (a) and Figure 7 As can be seen from the comparison in (b), setting the annular spacer 170 can effectively block the excitation light incident at a large angle, while improving the image clarity, especially optimizing the contrast between teeth and plaque in the image, increasing the color contrast between teeth and plaque, and thus improving the accuracy of subsequent image analysis.

[0095] Through the structural design of this embodiment, the dental plaque observer 100 can effectively improve the quality of dental fluorescence images and reduce abnormal brightness caused by excitation light reflection, thereby ensuring clearer imaging of the plaque area, significantly improving the quality of fluorescence images, and thus enhancing the accuracy of subsequent image analysis by the host computer.

[0096] In some specific embodiments, such as Figure 6 As shown, at least one side of the transparent window 150 is coated with a light-shielding layer 180 at a position opposite to the annular spacer 170. The light-shielding layer 180 is annular. Specifically, the central region of the transparent window 150 is aligned with the light-incident end of the optical lens assembly 120, and the outer ring region of the transparent window 150 is aligned with the light-emitting surface of the light source 130. The surfaces of these two regions are coated with a white light anti-reflection coating to improve light transmittance and maintain brightness in white light observation mode. A black light-shielding film is coated between the central region and the outer ring region to form an annular light-shielding layer 180. The light-shielding layer 180 can be coated on the inner surface of the transparent window 150 near the annular spacer 170, the outer surface on the opposite side, or both sides. The transmittance of the light-shielding layer 180 is less than 1%. The light-shielding layer 180 also corresponds to the aperture structure of the optical lens assembly, used to control light throughput, block unnecessary light, and further reduce interference from internal stray light, thereby improving the quality of the two-dimensional fluorescence image.

[0097] In some specific embodiments, the projection of the annular spacer 170 onto the transparent window 150 does not extend beyond the area where the light-shielding layer is located. The inner ring of the annular spacer 170 does not sacrifice the field of view of the lens, and the outer ring does not sacrifice the brightness of the light source. In one specific embodiment, the ring width of the annular spacer 170 is in the range of 1 ± 0.35 mm, and the ring width of the light-shielding layer 180 is slightly larger than that of the annular spacer 170. Because the distance between the light source 130 and the optical lens group 120 is relatively close, a small portion of the excitation light emitted by the light source 130 is reflected by the transparent window 150. The reflected light may enter the lens group at a large angle, resulting in abnormally high brightness. The light-shielding layer 180 and the annular spacer 170, both with a certain width, can block the excitation light incident at a large angle to a certain extent, preventing the reflected light from affecting the imaging of the fluorescence signal, thereby improving the quality and accuracy of the image.

[0098] In some specific embodiments, the host computer performs step S2, which involves image segmentation of the plaque region and the tooth region based on the two-dimensional fluorescence image of the teeth to obtain plaque segmentation results and tooth segmentation results, including:

[0099] The host computer calls a pre-trained dental plaque segmentation model to perform image segmentation on the two-dimensional fluorescence image of the teeth, and outputs the plaque segmentation result and the tooth segmentation result, such as... Figure 8 (a) is a two-dimensional fluorescence image of the teeth. Figure 8 (b) is the corresponding segmented image; wherein the dental plaque segmentation model is obtained through the following training method:

[0100] S201: Construct a training dataset and a validation dataset. Both the training dataset and the validation dataset contain several training images. Each training image is a two-dimensional fluorescence image of teeth, and each training image has been pre-segmented and labeled with the tooth region and the dental plaque region.

[0101] S202: Construct the initial segmentation model, which adopts the TransUNet architecture;

[0102] S203: Input the training images in the training dataset into the initial segmentation model to obtain the model's prediction results for dental plaque and tooth regions;

[0103] S204: Calculate the loss value between the model's prediction results and the labeled results, using the following loss function: ,in, and These are the weighting coefficients of the loss function. , , The number of dental plaque pixels correctly identified by the model. The number of non-dental plaque pixels that the model incorrectly identified as dental plaque. This represents the number of dental plaque pixels not identified by the model. This is the difficulty weighting coefficient. , This is the penalty coefficient;

[0104] S205: Update the parameters of the initial segmentation model iteratively using the backpropagation algorithm to minimize the loss value and obtain the transient segmentation model;

[0105] S206: During the training process, the transient segmentation model is used to test and verify the validation dataset. If the validation result meets the preset iteration cutoff condition, the validated transient segmentation model is used as the trained dental plaque segmentation model.

[0106] In this embodiment, in step S201 above, an image dataset and a validation dataset are constructed for training. Each training image is divided into a tooth region and a plaque region, and these regions are pre-labeled. The training dataset includes tooth images from different angles, containing plaque of different shapes and sizes, as well as different tooth image samples without plaque. A large number of images in the training and validation datasets are segmented and labeled using the X-AnyLabeling tool, and each image is segmented to obtain labels for the tooth region and the plaque region.

[0107] In step S202 above, TransUNet is used as the initial segmentation model. The PyTorch framework is used to construct the initial segmentation model, and it is trained on an RTX4090 GPU to accelerate computation and improve training efficiency. In the decoder of the initial segmentation model, the output of each upsampling module is connected in series with a spatial-channel attention fusion module to enhance the model's feature extraction capability. The spatial-channel attention fusion module (SCSE Attention, Spatial+ChannelSqueeze and Excitation) includes a Channel-SE submodule and a Spatial-SE submodule. The Channel-SE submodule is used to enhance the saliency of plaque region features, while the Spatial-SE submodule is used to strengthen the spatial distribution features of plaque regions. By fusing the output features of the two submodules, the segmentation model can better capture blurred edges and small target regions when processing plaque regions.

[0108] In step S203 above, images from the training dataset are input into the initial segmentation model for training. Each image undergoes forward propagation through a neural network to calculate the model's predictions for plaque and tooth regions, which are then compared with the actual annotations to calculate a loss value. The loss value is used to update the model weights through a backpropagation algorithm to progressively optimize the segmentation effect. This process is iterated until a preset training accuracy is reached; for example, training at epoch=200.

[0109] In step S204 above, during each training process, the model calculates the loss value based on the difference between the current prediction result and the actual annotation. The loss function is a weighted combination of Focal Tversky Loss and Dice Loss. and Here are the weighting coefficients of the loss function. In a specific embodiment, m = 0.7, n = 0.3, and the loss function formula is: α is the penalty coefficient for false positives (FP), used to control the emphasis on "false detections"; β is the penalty coefficient for false negatives (FN), used to control the emphasis on "false negatives" (FN). When β > α, the emphasis is on reducing false negatives (FN), suitable for situations with many small targets; when α > β, the emphasis is on reducing false positives (FP), suitable for situations with a lot of background. When γ > 1, "difficult-to-predict regions (low TL)" receive more attention. Preferably, γ = 2, which provides better stability. By optimizing the loss function, the segmentation model in this embodiment focuses on pixel recognition of blurry contours, low contrast, small-sized targets, and a small number of important dental plaque regions.

[0110] In steps S205-S206 above, the model parameters are adjusted according to the loss value, and the model is continuously iterated and optimized until the loss value converges and the preset accuracy requirement is reached, resulting in the final trained dental plaque segmentation model that can accurately identify dental plaque regions.

[0111] In some specific embodiments, the system further includes a display, which is communicatively connected to the host computer. The host computer performs step S4, which involves performing three-dimensional reconstruction based on the plaque segmentation results, tooth segmentation results, and the relative depth map to obtain a three-dimensional tooth model containing plaque and tooth regions. This step includes:

[0112] S401: The host computer constructs a point cloud map of the tooth surface based on the relative depth map;

[0113] S402: Based on the plaque segmentation results and the tooth segmentation results, different textures are applied to the point cloud map to process the mapping of different regions, thereby generating a three-dimensional tooth model that includes the plaque region and the tooth region.

[0114] S403: Display the three-dimensional model of the teeth on the monitor.

[0115] In this embodiment, in step S401 above, the relative depth map obtained by the depth estimation model is used to associate the pixel coordinates of the two-dimensional fluorescence image of the tooth with the depth information, thereby realizing the construction of a three-dimensional point cloud of the tooth surface. This embodiment is based on DepthAnything v2 for depth estimation. DepthAnything v2 uses high-quality synthetic data to train a teacher model, and then uses the teacher model to generate pseudo-depth labels from massive amounts of real data, and then trains a student model based on these data. It has fast streaming speed and strong generalization performance. Inference is performed based on the vitb model in the DepthAnything v2 variant, with the GPU configuration of RTX 4060, and the relative depth map of the two-dimensional image of the tooth is output. The relative depth map is single-channel, floating-point data, and the values ​​show the characteristic of being larger when closer and smaller when farther away. By mapping each pixel in the two-dimensional image with its relative depth value, a point cloud map of the tooth surface is generated. Each point in the point cloud map represents the coordinate information of the tooth surface in three-dimensional space, and the point cloud map contains the geometric shape and position data of the tooth surface.

[0116] In step S402 above, different regions in the point cloud are marked according to the segmentation results. Then, different textures are used for mapping different regions, so that the plaque region and the tooth region in the 3D model have obvious visual differences.

[0117] In step S403 above, the textured 3D model of the teeth is displayed on the connected monitor, such as... Figure 9 (a) Figure 9 (b) and Figure 9 As shown in (c), users can intuitively observe the three-dimensional morphology of the tooth surface and plaque from different angles, and understand the distribution of plaque on the tooth surface. The display can be interactive, allowing users to rotate and view the three-dimensional tooth model from different angles, thereby helping doctors or users to more accurately assess the condition of teeth and plaque.

[0118] In some specific embodiments, the host computer performs a step S5 of triangulating the surface of the three-dimensional tooth model, calculating the surface area of ​​the plaque region based on the triangulation results, and calculating the surface area of ​​the tooth region, including:

[0119] S501: The host computer extracts the corresponding surface point cloud from the three-dimensional tooth model according to the plaque segmentation result and the tooth segmentation result, respectively, to obtain the plaque cloud set and the tooth point cloud set;

[0120] S502: Perform Delaunay triangulation on the plaque cloud set and the tooth point cloud set respectively to form plaque region triangular mesh and tooth region triangular mesh respectively, wherein the plaque region triangular mesh and the tooth region triangular mesh each contain a number of triangular pieces;

[0121] S503: Determine the boundary point set based on the plaque segmentation result and the tooth segmentation result. Filter the triangular pieces in the triangular mesh of the plaque region and the triangular mesh of the tooth region respectively. Select the triangular pieces whose three vertices all belong to the boundary point set as hollow triangular pieces and discard the hollow triangular pieces.

[0122] S504: The areas of the plaque region triangles after removing the cavity triangles are summed to obtain the surface area of ​​the plaque region; and the areas of the tooth region triangles after removing the cavity triangles are summed to obtain the surface area of ​​the tooth region.

[0123] In this embodiment, in steps S501-S502 above, the plaque region point cloud set and the tooth region point cloud set respectively contain detailed spatial coordinate data of the tooth surface and the plaque region, which are used for calculation in subsequent steps. Through Delaunay triangulation, each target point in the plaque cloud set is connected to other points in its neighborhood to form several triangles. Finally, all the triangles in the plaque cloud set are combined into a plaque region triangular mesh; and each target point in the tooth point cloud set is connected to other points in its neighborhood to form several triangles. Finally, all the triangles in the tooth point cloud set are combined into a tooth region triangular mesh.

[0124] In step S503 above, the host computer determines the boundary point set based on the plaque segmentation results and the tooth segmentation results. The boundary point set refers to the points located at the boundary between the plaque and tooth regions. Specifically, a mask can be obtained from the segmentation results, and the OpenCV contour calculation method findContours is used to find contour pixels, which correspond to the boundary points of the point cloud. Based on this boundary point set, the host computer filters each triangle in the triangular mesh of the plaque region and the triangular mesh of the tooth region. If all three vertices of a triangle belong to the boundary point set, the triangle is considered an empty triangle and needs to be removed to improve the calculation accuracy.

[0125] In step S504 above, the host computer performs surface area summation on the plaque region triangles and tooth region triangles after removing the cavity triangles, ultimately obtaining the total surface area of ​​the plaque region and the total surface area of ​​the tooth region. A common method for calculating the area of ​​a triangle is to use the cross product of vectors.

[0126] In some specific embodiments, the host computer performs Delaunay triangulation on the plaque cloud set and the tooth point cloud set respectively, forming plaque region triangular meshes and tooth region triangular meshes respectively, including:

[0127] S5021: The host computer takes any target point in the bacterial spot cloud set and the tooth point cloud set as the center, and selects a preset number of neighboring points around the target point through the K-nearest neighbor algorithm to form the neighborhood point set of the center.

[0128] S5022: The coordinates of the neighborhood point set are subtracted from the coordinates of the target point for centering, and a covariance matrix is ​​constructed.

[0129] S5023: Solve for the eigenvector corresponding to the smallest eigenvalue of the covariance matrix through principal component analysis, and use it as the normal vector of the target point;

[0130] S5024: For any two target points to be connected, obtain the normal vectors corresponding to them respectively, and calculate the spatial angle between the two normal vectors;

[0131] S5025: Determine whether the spatial angle is less than or equal to a preset angle threshold;

[0132] S5026: If so, connect the two target points to be connected to form the effective edge of the triangulation;

[0133] S5027: Based on all valid edges connected in the set of bacterial spot clouds, form a triangular mesh for the bacterial spot region; and based on all valid edges connected in the set of tooth point clouds, form a triangular mesh for the tooth region.

[0134] In this embodiment, for any target point in the bacterial spot cloud or tooth point cloud, the K-nearest neighbor algorithm is invoked to select a predetermined number (e.g., k=8) of neighboring points around the target point. Using the target point's 3D coordinates as a reference, the 8 points with the smallest distance are selected to form the target point's neighborhood point set. The 3D coordinates of each point in the neighborhood point set are subtracted from the target point's 3D coordinates to obtain the centered coordinates. Based on the centered neighborhood point set, a 3×3 covariance matrix C is constructed. Principal component analysis (PCA) is performed on the covariance matrix C to obtain the three eigenvalues ​​and corresponding eigenvectors of the covariance matrix through eigenvalue decomposition. The eigenvector corresponding to the smallest eigenvalue is selected as the normal vector of the current target point. This normal vector represents the surface orientation of the local region where the target point is located. For any two target points to be connected, their corresponding normal vectors are obtained, and the spatial angle θ between the two normal vectors is calculated using the vector dot product formula. The system determines whether the spatial angle θ is less than or equal to a preset angle threshold. In one specific embodiment, the preset angle threshold is set to 30°. This angle is set based on the special curvature structure of the tooth surface morphology, mainly to prevent the misinterpretation of dental plaque with concave surfaces formed in the gaps between teeth. Setting the angle too large may lead to incorrect connections and the generation of redundant triangles, while setting the angle too small may lead to the over-removal of some areas that should be connected. If θ ≤ the preset angle threshold, for example, 30°, it is determined that the local curved surfaces of the two target points have the same orientation, and the two are connected to form an effective edge of the triangulation; if θ > the preset angle threshold, for example, 30°, the two target points are not connected.

[0135] In some optional embodiments, a preset angle threshold is dynamically adjusted based on the shape, size, and / or curvature corresponding to different tooth positions. For example, different preset angle thresholds can be set for incisors, canines, and molars. The specific value can be adjusted up or down (e.g., ±5°) from the 30° selected in the above embodiments to adapt to different tooth morphologies, thereby further improving the triangulation accuracy of concave plaque areas and concave tooth areas.

[0136] Dental plaque easily forms in the gaps between teeth, especially in the concave areas between adjacent teeth, often forming plaque clusters with complex shapes. This embodiment calculates the spatial angle between the normal vectors of two target points to be connected and filters valid edges using a preset angle threshold (e.g., 30°). This effectively filters out invalid connections between points with significantly different surface orientations, particularly for points like... Figure 10 The plaque that forms in the concave areas, such as the gap between two teeth, as shown in the image, helps prevent plaque buildup caused by incorrect connections, such as... Figure 11 The triangulation distortion shown results in the following: Figure 12 The correct concave surface subdivision results shown ensure the accuracy of subsequent surface area calculations.

[0137] like Figure 2 As shown, in one embodiment, a method for quantitative analysis of dental plaque is provided. This method is executed by a host computer and includes:

[0138] S1: Receive two-dimensional fluorescent images of teeth sent by the dental plaque detector;

[0139] S2: Based on the two-dimensional fluorescence image of the teeth, perform image segmentation on the plaque region and the tooth region to obtain plaque segmentation results and tooth segmentation results;

[0140] S3: Perform two-dimensional image depth estimation on the two-dimensional fluorescence image of the tooth to obtain a relative depth map used to characterize the spatial information of the tooth surface;

[0141] S4: Perform three-dimensional reconstruction based on the plaque segmentation results, tooth segmentation results, and relative depth map to obtain a three-dimensional tooth model containing plaque and tooth regions;

[0142] S5: Triangulate the surface of the three-dimensional tooth model, calculate the surface area of ​​the plaque region based on the triangulation results, calculate the surface area of ​​the tooth region, and obtain the area ratio of the surface area of ​​the plaque region to the surface area of ​​the tooth region. Use the area ratio as the quantitative analysis result of dental plaque.

[0143] In some specific embodiments, step S2, which involves image segmentation of the plaque region and the tooth region based on the two-dimensional fluorescence image of the teeth to obtain plaque segmentation results and tooth segmentation results, includes:

[0144] The pre-trained dental plaque segmentation model is used to segment the two-dimensional fluorescence image of the teeth, outputting plaque segmentation results and tooth segmentation results; wherein, the dental plaque segmentation model is obtained through the following training method:

[0145] A training dataset and a validation dataset are constructed. Each training dataset contains several training images. Each training image is a two-dimensional fluorescence image of teeth, and each training image has been pre-segmented and labeled with the tooth region and the dental plaque region.

[0146] An initial segmentation model is constructed, which adopts the TransUNet architecture;

[0147] The training images in the training dataset are input into the initial segmentation model to obtain the model's prediction results for dental plaque and tooth regions.

[0148] The loss value between the model's prediction results and the labeled results is calculated using the following loss function: ,in, and These are the weighting coefficients of the loss function. , , The number of dental plaque pixels correctly identified by the model. The number of non-dental plaque pixels that the model incorrectly identified as dental plaque. This represents the number of dental plaque pixels not identified by the model. This is the difficulty weighting coefficient. , This is the penalty coefficient;

[0149] The parameters of the initial segmentation model are updated iteratively using the backpropagation algorithm to minimize the loss value, thereby obtaining a transient segmentation model.

[0150] During training, the transient segmentation model is used to test and verify the validation dataset. If the validation result meets the preset iteration cutoff condition, the validated transient segmentation model is used as the trained dental plaque segmentation model.

[0151] In some specific embodiments, step S4, which involves performing three-dimensional reconstruction based on the plaque segmentation results, tooth segmentation results, and the relative depth map to obtain a three-dimensional tooth model containing both plaque and tooth regions, includes:

[0152] Based on the relative depth map, a point cloud map of the tooth surface is constructed;

[0153] Based on the plaque segmentation results and the tooth segmentation results, different textures are applied to the corresponding regions in the point cloud image to generate a three-dimensional tooth model containing plaque regions and tooth regions.

[0154] The three-dimensional model of the teeth is displayed on the monitor.

[0155] In some specific embodiments, step S5, which involves triangulating the surface of the three-dimensional tooth model, calculating the surface area of ​​the plaque region based on the triangulation results, and calculating the surface area of ​​the tooth region, includes:

[0156] Based on the plaque segmentation results and the tooth segmentation results, the corresponding surface point clouds are extracted from the three-dimensional tooth model to obtain the plaque cloud set and the tooth point cloud set.

[0157] Delaunay triangulation is performed on the bacterial spot cloud set and the tooth point cloud set respectively to form a bacterial spot region triangular mesh and a tooth region triangular mesh respectively, wherein the bacterial spot region triangular mesh and the tooth region triangular mesh each contain a number of triangular pieces;

[0158] Based on the plaque segmentation results and the tooth segmentation results, the boundary point set is determined. The triangular pieces in the triangular mesh of the plaque region and the triangular mesh of the tooth region are screened respectively. The triangular pieces whose three vertices all belong to the boundary point set are regarded as hollow triangular pieces and the hollow triangular pieces are removed.

[0159] The surface area of ​​the plaque region is obtained by summing the areas of the triangular pieces in the plaque region after removing the cavities; and the surface area of ​​the tooth region is obtained by summing the areas of the triangular pieces in the tooth region after removing the cavities.

[0160] In some specific embodiments, the steps of performing Delaunay triangulation on the plaque cloud set and the tooth point cloud set to form plaque region triangular meshes and tooth region triangular meshes respectively include:

[0161] Taking any target point in the bacterial spot cloud set and the tooth point cloud set as the center, a preset number of neighboring points around the target point are selected by the K-nearest neighbor algorithm to form the neighborhood point set of the center.

[0162] The coordinates of the neighborhood point set are subtracted from the coordinates of the target point for centering, and a covariance matrix is ​​constructed.

[0163] The eigenvector corresponding to the smallest eigenvalue of the covariance matrix is ​​obtained by principal component analysis and used as the normal vector of the target point.

[0164] For any two target points to be connected, obtain their corresponding normal vectors and calculate the spatial angle between the two normal vectors.

[0165] Determine whether the spatial angle is less than or equal to a preset angle threshold;

[0166] If so, connect the two target points to be connected to form the effective edges of the triangulation;

[0167] Based on all valid edges connected in the set of bacterial spot clouds, a triangular mesh for the bacterial spot region is formed; and based on all valid edges connected in the set of tooth point clouds, a triangular mesh for the tooth region is formed.

[0168] Figure 13 An internal structural diagram of a computer device in one embodiment is shown. This computer device can specifically be a terminal or a server. Figure 13As shown, the computer device includes a processor, memory, and network interface connected via a system bus. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and may also store a computer program. When executed by the processor, this computer program enables the processor to implement a method for quantitative analysis of dental plaque. The internal memory may also store a computer program, which, when executed by the processor, enables the processor to implement the method for quantitative analysis of dental plaque. Those skilled in the art will understand that… Figure 13 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0169] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0170] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0171] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A system for quantitative analysis of dental plaque, characterized in that, The system includes a dental plaque observation instrument and a host computer. The dental plaque observation instrument is used to acquire two-dimensional fluorescence images of teeth in the oral cavity area of ​​the subject. The host computer is communicatively connected to the dental plaque observation instrument and is configured as follows: Receive two-dimensional fluorescence images of teeth transmitted by the dental plaque observation instrument; Based on the two-dimensional fluorescence image of the teeth, the plaque region and the tooth region are segmented to obtain the plaque segmentation result and the tooth segmentation result. Two-dimensional image depth estimation is performed on the two-dimensional fluorescence image of the tooth to obtain a relative depth map that characterizes the spatial information of the tooth surface. Based on the plaque segmentation results, tooth segmentation results, and relative depth map, a three-dimensional reconstruction is performed to obtain a three-dimensional tooth model containing plaque and tooth regions. The surface of the three-dimensional tooth model is triangulated, and the surface area of ​​the plaque region is calculated based on the triangulation results. The surface area of ​​the tooth region is also calculated, and the area ratio of the surface area of ​​the plaque region to that of the tooth region is obtained. The area ratio is used as the quantitative analysis result of dental plaque. The surface area of ​​the plaque region is calculated by summing the areas of the triangular pieces of the plaque region after removing the cavity triangular pieces.

2. The dental plaque quantitative analysis system according to claim 1, characterized in that, The dental plaque observation instrument includes an observation instrument body, an optical lens group, a light source, an image sensor, a transparent window, a filter, and an annular spacer. The light source is configured to switch between ultraviolet light and white light. The main body of the observation instrument has an opening, the optical lens group is disposed within the opening, the light source is disposed around the light-incident end of the optical lens group, the image sensor is disposed at the rear end of the optical lens group, a transparent window is disposed at the opening of the opening, the transparent window is sealed to the main body of the observation instrument, the transparent window covers at least the light-incident end of the optical lens group and the light-emitting surface of the light source, and a filter is disposed between the transparent window and the image sensor; An annular spacer is provided between the transparent window and the optical lens assembly. One side of the annular spacer abuts against the surface of the transparent window and the other side of the annular spacer abuts against the surface of the optical lens assembly. The annular spacer is made of an opaque material.

3. The dental plaque quantitative analysis system according to claim 2, characterized in that, At least one side of the transparent window sheet is coated with a light-shielding layer at a position opposite to the annular spacer, and the light-shielding layer is annular.

4. The dental plaque quantitative analysis system according to claim 3, characterized in that, The projection of the annular spacer onto the transparent window does not extend beyond the area where the light-shielding layer is located.

5. The dental plaque quantitative analysis system according to claim 1, characterized in that, The host computer executes the step of image segmentation of the plaque region and the tooth region based on the two-dimensional fluorescence image of the tooth, to obtain the plaque segmentation result and the tooth segmentation result, including: The host computer calls a pre-trained dental plaque segmentation model to perform image segmentation on the two-dimensional fluorescence image of the teeth, and outputs plaque segmentation results and tooth segmentation results; wherein, the dental plaque segmentation model is obtained through the following training method: A training dataset and a validation dataset are constructed. Each training dataset contains several training images. Each training image is a two-dimensional fluorescence image of teeth, and each training image has been pre-segmented and labeled with the tooth region and the dental plaque region. An initial segmentation model is constructed, which adopts the TransUNet architecture; The training images in the training dataset are input into the initial segmentation model to obtain the model's prediction results for dental plaque and tooth regions. The loss value between the model's prediction results and the labeled results is calculated using the following loss function: ,in, and These are the weighting coefficients of the loss function. , , The number of dental plaque pixels correctly identified by the model. The number of non-dental plaque pixels that the model incorrectly identified as dental plaque. This represents the number of dental plaque pixels not identified by the model. This is the difficulty weighting coefficient. , This is the penalty coefficient; The parameters of the initial segmentation model are updated iteratively using the backpropagation algorithm to minimize the loss value, thereby obtaining a transient segmentation model. During training, the transient segmentation model is used to test and verify the validation dataset. If the validation result meets the preset iteration cutoff condition, the validated transient segmentation model is used as the trained dental plaque segmentation model.

6. The dental plaque quantitative analysis system according to claim 1, characterized in that, The system also includes a display, which is communicatively connected to the host computer. The host computer performs a step of three-dimensional reconstruction based on the plaque segmentation results, tooth segmentation results, and the relative depth map to obtain a three-dimensional tooth model containing plaque and tooth regions, including: The host computer constructs a point cloud map of the tooth surface based on the relative depth map; Based on the plaque segmentation results and the tooth segmentation results, different textures are applied to the corresponding regions in the point cloud image to generate a three-dimensional tooth model containing plaque regions and tooth regions. The three-dimensional model of the teeth is displayed on the monitor.

7. The dental plaque quantitative analysis system according to claim 1, characterized in that, The host computer performs the following steps: triangulation of the surface of the three-dimensional tooth model, calculation of the surface area of ​​the plaque region based on the triangulation results, and calculation of the surface area of ​​the tooth region. The host computer extracts the corresponding surface point clouds from the three-dimensional tooth model based on the plaque segmentation results and the tooth segmentation results, respectively, to obtain a plaque cloud set and a tooth point cloud set. Delaunay triangulation is performed on the bacterial spot cloud set and the tooth point cloud set respectively to form a bacterial spot region triangular mesh and a tooth region triangular mesh respectively, wherein the bacterial spot region triangular mesh and the tooth region triangular mesh each contain a number of triangular pieces; Based on the plaque segmentation results and the tooth segmentation results, the boundary point set is determined. The triangular pieces in the triangular mesh of the plaque region and the triangular mesh of the tooth region are screened respectively. The triangular pieces whose three vertices all belong to the boundary point set are regarded as hollow triangular pieces and the hollow triangular pieces are removed. The surface area of ​​the plaque region is obtained by summing the areas of the triangular pieces in the plaque region after removing the cavities; and the surface area of ​​the tooth region is obtained by summing the areas of the triangular pieces in the tooth region after removing the cavities.

8. The dental plaque quantitative analysis system according to claim 7, characterized in that, The host computer performs Delaunay triangulation on the plaque cloud set and the tooth point cloud set respectively, forming plaque region triangular meshes and tooth region triangular meshes respectively, including: The host computer takes any target point in the bacterial spot cloud set and the tooth point cloud set as the center, and selects a preset number of neighboring points around the target point using the K-nearest neighbor algorithm to form the neighborhood point set of the center. The coordinates of the neighborhood point set are subtracted from the coordinates of the target point for centering, and a covariance matrix is ​​constructed. The eigenvector corresponding to the smallest eigenvalue of the covariance matrix is ​​obtained by principal component analysis and used as the normal vector of the target point. For any two target points to be connected, obtain their corresponding normal vectors and calculate the spatial angle between the two normal vectors. Determine whether the spatial angle is less than or equal to a preset angle threshold; If so, connect the two target points to be connected to form the effective edges of the triangulation; Based on all valid edges connected in the set of bacterial spot clouds, a triangular mesh for the bacterial spot region is formed; and based on all valid edges connected in the set of tooth point clouds, a triangular mesh for the tooth region is formed.

9. A method for quantitative analysis of dental plaque, characterized in that, The dental plaque quantitative analysis system as described in any one of claims 1-8, wherein the dental plaque quantitative analysis method comprises: Receive two-dimensional fluorescent images of teeth sent by a dental plaque detector; Based on the two-dimensional fluorescence image of the teeth, the plaque region and the tooth region are segmented to obtain the plaque segmentation result and the tooth segmentation result. Two-dimensional image depth estimation is performed on the two-dimensional fluorescence image of the tooth to obtain a relative depth map that characterizes the spatial information of the tooth surface. Based on the plaque segmentation results, tooth segmentation results, and relative depth map, a three-dimensional reconstruction is performed to obtain a three-dimensional tooth model containing plaque and tooth regions. The surface of the three-dimensional tooth model is triangulated, and the surface area of ​​the plaque region is calculated based on the triangulation results. The surface area of ​​the tooth region is also calculated, and the area ratio of the surface area of ​​the plaque region to that of the tooth region is obtained. The area ratio is used as the quantitative analysis result of dental plaque. The surface area of ​​the plaque region is calculated by summing the areas of the triangular pieces of the plaque region after removing the cavity triangular pieces.

10. A computer-readable storage medium, characterized in that, The device contains a computer program that, when executed by a processor, causes the processor to perform the steps of the method for quantitative analysis of dental plaque as described in claim 9.

Images