Oral cavity examination method and system based on big data analysis

Abstract

The invention belongs to the technical field of oral health detection, and discloses an oral examination method and system based on big data analysis, and the method comprises the steps: collecting oral image data and a corresponding image evaluation label, carrying out the distortion conversion of the oral image data, and obtaining an oral correction image; performing reflection compensation on the oral cavity correction image to obtain an oral cavity compensation image; performing lesion segmentation on the oral cavity compensation map to obtain lesion sub-maps; an oral health assessment model is constructed based on the lesion sub-graphs and the corresponding image assessment labels, and accurate assessment of oral health is realized based on the oral health assessment model; the accuracy of oral health assessment of different patients is greatly improved, and implementation of personalized oral care and health management is further promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of oral health detection. More specifically, the present invention relates to an oral examination method and system based on big data analysis. Background Art

[0002] With the rapid development of modern oral medicine, the importance of oral examination in disease prevention, early diagnosis, and personalized treatment has become increasingly prominent. However, existing oral examination technologies still face many technical problems. For example, the inner surface of the oral cavity has complex curved surfaces and irregular shapes, including the natural curvature and folding of parts such as teeth, tongue, and gums. These structures cause severe distortion of the images during imaging, making it difficult to accurately restore the position and shape of the lesion area, resulting in low calibration accuracy and difficulty in meeting clinical requirements. During the oral examination process, the presence of saliva is inevitable. The surface of saliva has strong light reflection characteristics. Especially under high-brightness light sources, it is easy to generate strong reflective areas in imaging. These reflective areas will cover up the detailed information of the lesion area and reduce the recognition rate of the lesion. The color and texture differences of oral tissues are small, especially the differences in imaging characteristics between the lesion area and the surrounding healthy tissues are not obvious. Due to the blurred boundaries, traditional edge detection or simple threshold segmentation methods often have difficulty effectively segmenting the lesion area. In addition, complex lesions may overlap with each other, further increasing the segmentation difficulty.

[0003] In view of this, the present invention proposes an oral examination method and system based on big data analysis to solve the above problems. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art and to achieve the above object, the present invention provides the following technical solution: An oral examination method based on big data analysis, characterized by comprising:

[0005] S1. Collect oral image data and corresponding image evaluation labels, perform distortion conversion on the oral image data to obtain an oral correction map;

[0006] S2. Perform reflection compensation on the oral correction image to obtain an oral compensation map;

[0007] S3. Perform lesion segmentation on the oral compensation map to obtain a lesion sub-map;

[0008] S4. Construct an oral health assessment model based on the lesion sub-map and the corresponding image evaluation labels, and achieve accurate assessment of oral health based on the oral health assessment model.

[0009] Furthermore, the method for obtaining the oral image data includes: using a high-definition oral endoscope equipped with a multi-angle imaging function and a macro lens, setting the angle and position of the device, combining with a flexible arm mechanical assistance system, and using the geometric registration technology of multi-view images to splice different-view images. The generated panoramic oral image is the oral image data.

[0010] Furthermore, the method for performing distortion conversion on the oral image data includes:

[0011] The oral image data is grayscaled using the RGB weighted average method to obtain an oral grayscale image; for each pixel in the oral grayscale image, the pixel in the lower left corner of the oral grayscale image is used as the pixel center, and the pixel height and pixel width of the oral grayscale image are used as the coordinate scales to assign coordinates to each pixel, obtaining the pixel coordinates of each pixel; the horizontal gradient operator in the edge detection operator is used to calculate the horizontal gradient of the pixels in the oral grayscale image to obtain the horizontal pixel gradient, and the vertical gradient operator in the edge detection operator is used to calculate the vertical gradient of the pixels in the oral grayscale image to obtain the vertical pixel gradient; a pixel tensor of each pixel is constructed based on the horizontal pixel gradient and the vertical pixel gradient, and distortion evaluation is performed on each pixel based on the pixel tensor to obtain a distortion estimate value;

[0012] A preset distortion threshold is set, and the pixels with distortion estimate values greater than or equal to the distortion threshold are marked as distorted pixels; distortion displacement exploration is performed on each pixel marked as a distorted pixel to obtain a distortion displacement field; pixel transfer is performed on the distorted pixels based on the distortion displacement field, the horizontal distortion offset and the vertical distortion offset in the distortion displacement field are used to calculate the coordinates of the distorted pixels to obtain offset coordinates, the distorted pixels are replaced with the pixels at the offset coordinates, the pixels directly adjacent to the distorted pixels are used as neighborhood pixels, and the pixel values at the distorted pixels are filled with the pixel mean values of the neighborhood pixels to obtain an oral correction image.

[0013] Furthermore, the formula for performing distortion evaluation on each pixel is:

[0014] where AS(x,y) represents the distortion estimate value of the pixel at pixel coordinates (x,y), |ZL(x,y)| represents the modulus of the pixel tensor of the pixel at pixel coordinates (x,y), Gx represents the horizontal pixel gradient, Gy represents the vertical pixel gradient, x represents the horizontal axis value of the pixel coordinates, and y represents the vertical axis value of the pixel coordinates.

[0015] Furthermore, the method for performing distortion displacement exploration on each pixel marked as a distorted pixel includes:

[0016] Initialize the distortion offset group of Group A. Initialize the best offset group as a null value. The distortion offset group includes horizontal distortion offset and vertical distortion offset. Based on the distortion offset group, construct a distortion energy field function. The expression of the distortion energy field function is as follows:

[0017] where E represents the distortion energy, α1 represents the horizontal weight coefficient, represents the horizontal offset gradient, α2 represents the vertical weight coefficient, represents the vertical offset gradient, β represents the pixel distortion parameter, I(x, y) represents the pixel grayscale of the pixel at pixel coordinates (x, y), I(x + u, y + v) represents the pixel grayscale of the pixel at pixel coordinates (x + u, y + v), u represents the horizontal distortion offset, v represents the vertical distortion offset, dx represents the integration of the horizontal axis value, and dy represents the integration of the vertical axis value; Calculate the distortion energy of each group of distortion offset groups based on the distortion energy field function. Based on the distortion energy, use the tournament selection algorithm to initially select the distortion offset groups to obtain the initial distortion offset groups;

[0018] Pair the initial distortion offset groups in pairs to obtain offset pairings. Recombine the horizontal distortion offset and vertical distortion offset in each offset pairing to obtain offspring combinations different from the initial distortion offset groups; Calculate the distortion energy of each offspring combination through the distortion energy field function. Take the offspring combinations with distortion energy less than that of the offset pairings as the new generation; Preset a mutation scale. The mutation scale is a positive integer. Based on the mutation scale, randomly mutate the horizontal distortion offset or vertical distortion offset of the new generation. Randomly increase or decrease the horizontal distortion offset by one mutation scale, and randomly increase or decrease the vertical distortion offset by one mutation scale to obtain mutated combinations; Calculate the distortion energy of each mutated combination through the distortion energy field function. Take the mutated combinations with distortion energy less than that of the new generation as the preferred mutations. Select the preferred mutation with the minimum distortion energy as the candidate combination. When the best offset group is a null value, use the candidate combination as the new best offset group. When the best offset group is not a null value, if the distortion energy of the candidate combination is less than that of the best offset group, use the candidate combination as the new best offset group; Use the preferred mutation as the new initial distortion offset group and repeat until the distortion energy of the best offset group converges. Output the best offset group at this time as the distortion displacement field.

[0019] Furthermore, the method for performing specular reflection compensation on the oral cavity correction image includes:

[0020] Preset a set of scale components. Based on the set of scale components, perform light estimation on each pixel in the oral cavity correction image. The formula for performing light estimation on each pixel in the oral cavity correction image is: Among them, L(x, y) represents the illumination component of the pixel at pixel coordinates (x, y), CD represents the size of the scale component set, and w β represents the component weight of the β-th scale component in the scale component set. The component weight satisfies the component constraint condition, and the component constraint condition is: c β represents the β-th scale component in the scale component set, and F(c β , x, y) represents the convolution function; the reflection evaluation is performed on each pixel in the oral cavity correction image based on the illumination component to obtain the reflection component; the adaptive compensation estimation is performed on each pixel in the oral cavity correction image based on the reflection component to obtain the compensation factor; the light compensation is performed on each pixel in the oral cavity correction image based on the compensation factor to obtain the compensated pixel, and all the compensated pixels form the oral cavity compensation map.

[0021] Furthermore, the formula for performing reflection evaluation on each pixel in the oral cavity correction image is: Among them, R(x, y) represents the reflection component of the pixel at pixel coordinates (x, y), I(x, y) represents the pixel gray level of the pixel at pixel coordinates (x, y), and τ represents the offset constant;

[0022] The formula for performing adaptive compensation estimation on each pixel in the oral cavity correction image is: Among them, CH(x, y) represents the compensation factor of the pixel at pixel coordinates (x, y), γ represents the compensation control coefficient, represents the reflection gradient of the reflection component of the pixel at pixel coordinates (x, y).

[0023] Furthermore, the method for performing lesion segmentation on the oral cavity compensation map includes:

[0024] A preset pixel search radius is set. For the pixel B in the oral cavity compensation map, the Euclidean distance formula is used to calculate the pixel distance between other pixels and the pixel B, and the other pixels with a coordinate distance less than the pixel search radius from the pixel B form the pixel neighborhood set of the pixel B; the density evaluation is performed on each pixel in the oral cavity compensation map based on the pixel neighborhood to obtain the pixel gray level density; the distance measurement is performed on the pixels in the oral cavity compensation map based on the pixel gray level density to obtain the pixel distance;

[0025] Based on the pixel gray density, a clustering algorithm is used to cluster the pixels in the oral cavity compensation map to obtain pixel density clustering clusters; a density threshold is preset, the density mean value of each pixel density clustering cluster is calculated, and the pixel density clustering clusters with a density mean value greater than the density threshold are marked as density lesion regions; based on the pixel distance, a clustering algorithm is used to cluster the pixels in the oral cavity compensation map to obtain pixel distance clustering clusters; a distance threshold is preset, the distance mean value of each pixel density clustering cluster is calculated, and the pixel distance clustering clusters with a distance mean value greater than the distance threshold are marked as distance lesion regions; the density lesion regions and the distance lesion regions are compared for regional overlap, the pixels that are both density lesion regions and distance lesion regions are selected as lesion pixels, and the regions where the lesion pixels in the oral cavity compensation map are aggregated are cropped to obtain lesion sub - maps.

[0026] Further, the formula for evaluating the gray density of each pixel in the oral cavity compensation map is: where MD a represents the pixel gray density of the ath pixel in the oral cavity compensation map, Size represents the size of the pixel neighborhood set, I a represents the pixel gray of the ath pixel in the oral cavity compensation map, I j represents the pixel gray of the jth pixel in the pixel neighborhood set, θ represents the gray control parameter, d aj represents the pixel distance between the ath pixel and the jth pixel in the pixel neighborhood set in the oral cavity compensation map, and r represents the pixel search radius;

[0027] The formula for measuring the distance of pixels in the oral cavity compensation map is:

[0028] where Dis a represents the pixel distance of the ath pixel in the oral cavity compensation map, MD a represents the pixel gray density of the ath pixel in the oral cavity compensation map, MD j represents the pixel gray density of the jth pixel in the pixel neighborhood set, represents the density weight, JG(MD a , MD j ) represents the filtering function.

[0029] An oral cavity examination system based on big data analysis, comprising:

[0030] Data acquisition and processing module: Collect oral cavity image data and corresponding image evaluation labels, perform distortion conversion on the oral cavity image data to obtain an oral cavity correction map;

[0031] Image compensation module: Perform specular reflection compensation on the oral cavity correction image to obtain an oral cavity compensation map;

[0032] Region segmentation module: perform lesion segmentation on the oral cavity compensation map to obtain lesion sub - images;

[0033] Model construction module: construct an oral health assessment model based on the lesion sub - images and the corresponding image evaluation labels, and achieve accurate assessment of oral health based on the oral health assessment model.

[0034] The technical effects and advantages of an oral examination method and system based on big data analysis according to the present invention:

[0035] By performing distortion conversion on oral image data, the present invention effectively identifies and corrects the distorted areas in oral images, ensuring the accuracy of the images and the retention of details. Especially when dealing with corners, curved surfaces, or reflective areas inside the oral cavity, the influence of deformation is effectively reduced, improving the image quality. By performing specular reflection compensation on the corrected oral images, the image deviation caused by uneven illumination can be eliminated. Especially under high - brightness light sources, the saliva surface has strong light reflection characteristics, which easily generates strong reflective areas in imaging. The compensated image can better display the lesion area, making the subsequent lesion recognition more accurate. By performing lesion segmentation on the oral cavity compensation map, the lesion areas in the image are effectively identified, and the non - lesion areas are effectively filtered out, improving the accuracy of segmentation, precisely locating the lesion area, and reducing human error. By combining big data analysis and deep - learning models, the oral health of different patients can be accurately evaluated, thus promoting the realization of personalized oral care and health management. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic diagram of an oral examination method based on big data analysis according to the present invention;

[0037] Figure 2 It is a schematic diagram of an oral examination system based on big data analysis according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0039] Embodiment 1

[0040] Please refer to Figure 1 As shown, an oral examination method based on big data analysis described in this embodiment includes:

[0041] S1. Collect oral image data and corresponding image evaluation labels, perform distortion conversion on the oral image data to obtain an oral correction map;

[0042] S2. Perform specular reflection compensation on the oral correction image to obtain an oral compensation map;

[0043] S3. Perform lesion segmentation on the oral compensation map to obtain a lesion sub-map;

[0044] S4. Construct an oral health assessment model based on the lesion sub-map and the corresponding image evaluation labels, and achieve accurate assessment of oral health based on the oral health assessment model.

[0045] The acquisition method of oral image data includes: using a high-definition oral endoscope, equipped with multi-angle imaging function and macro lens, setting the angle and position of the device, combining with a flexible arm mechanical assistance system to make the image cover the key details of different areas of the oral cavity, using the geometric registration technology of multi-view images to splice different-view images, and the generated panoramic oral map is the oral image data;

[0046] The image evaluation label is the evaluation information of the oral image data, such as: for the gums, gum health, gum bleeding, gum inflammation, etc.; for the oral cavity, initial dental caries (early tooth cavity marks), severe dental caries (deep tooth cavities or decay), periodontal health, periodontitis, periodontal injury, etc.

[0047] The methods for performing distortion conversion on the oral image data include:

[0048] Use the RGB weighted average method to grayscale the oral image data to obtain an oral grayscale map; for each pixel in the oral grayscale map, use the pixel in the lower left corner of the oral grayscale map as the pixel center, and use the pixel height and pixel width of the oral grayscale map as the coordinate scales to assign coordinates to each pixel (for each additional pixel horizontally, the horizontal axis value of the pixel increases by one scale, and for each additional pixel vertically, the vertical axis value of the pixel increases by one scale) to obtain the pixel coordinates of each pixel; use the horizontal gradient operator in the edge detection operator to perform horizontal gradient calculation on the pixels in the oral grayscale map to obtain the horizontal pixel gradient, use the vertical gradient operator in the edge detection operator to perform vertical gradient calculation on the pixels in the oral grayscale map to obtain the vertical pixel gradient, and common edge detection operators include the Sobel operator and the Prewitt operator; construct the pixel tensor of each pixel based on the horizontal pixel gradient and the vertical pixel gradient; the expression of the pixel tensor is where, ZL(x,y) represents the pixel tensor of the pixel at pixel coordinates (x,y), Gx represents the horizontal pixel gradient, Gy represents the vertical pixel gradient, x represents the horizontal axis value of the pixel coordinates, and y represents the vertical axis value of the pixel coordinates; perform distortion evaluation on each pixel based on the pixel tensor, and the formula for performing distortion evaluation on each pixel is:

[0049] Among them, AS(x, y) represents the distortion estimation value of the pixel at pixel coordinates (x, y), and |ZL(x, y)| represents the modulus of the pixel tensor of the pixel at pixel coordinates (x, y). By performing distortion evaluation on each pixel, the distortion estimation value of each pixel is obtained. Based on the distortion estimation value, the distortion area in the image can be identified. The larger the distortion estimation value of the pixel, the stronger the deformation of the pixel. For example, in the corner, curved surface or reflective area of the oral cavity image, its local structure may cause a large gradient change, resulting in a large distortion estimation value;

[0050] A preset distortion threshold is set by those skilled in the art based on the actual situation; based on the distortion threshold, the pixels with distortion estimation values greater than or equal to the distortion threshold are marked as distorted pixels; for each pixel marked as a distorted pixel, a distorted displacement search is performed. Initialize group A of distortion offset groups, and initialize the best offset group as a null value. The distortion offset group includes horizontal distortion offset and vertical distortion offset. Based on the distortion offset group, a distortion energy field function is constructed. The expression of the distortion energy field function is: Among them, E represents the distortion energy, α1 represents the horizontal weight coefficient, which is used to control the influence degree of horizontal direction smoothness, represents the horizontal offset gradient, α2 represents the vertical weight coefficient, which is used to control the influence degree of vertical direction smoothness, represents the vertical offset gradient, β represents the pixel distortion parameter, which is used to control the influence of pixel error, I(x, y) represents the pixel gray level of the pixel at pixel coordinates (x, y), I(x + u, y + v) represents the pixel gray level of the pixel at pixel coordinates (x + u, y + v), u represents the horizontal distortion offset, v represents the vertical distortion offset, dx represents the integration of the horizontal axis value, and dy represents the integration of the vertical axis value;

[0051] Calculate the distortion energy of each group of distortion offsets based on the distortion energy field function, and initially select the distortion offset groups using the tournament selection algorithm based on the distortion energy to obtain the initial distortion offset groups; combine the initial distortion offset groups pairwise to obtain offset pairings, recombine the horizontal distortion offsets and vertical distortion offsets in each offset pairing to obtain offspring combinations different from the initial distortion offset groups; calculate the distortion energy of each offspring combination through the distortion energy field function, and use the offspring combinations with distortion energy less than that of the offset pairings as the new generation; preset a mutation scale, where the mutation scale is an integer greater than zero, randomly mutate the horizontal distortion offsets or vertical distortion offsets of the new generation based on the mutation scale, randomly increase or decrease a mutation scale for the horizontal distortion offsets, and randomly increase or decrease a mutation scale for the vertical distortion offsets to obtain mutated combinations; calculate the distortion energy of each mutated combination through the distortion energy field function, use the mutated combinations with distortion energy less than that of the new generation as the preferred mutations, select the preferred mutation with the minimum distortion energy as the candidate combination, when the best offset group is a null value, use the candidate combination as the new best offset group, and when the best offset group is not a null value, if the distortion energy of the candidate combination is less than that of the best offset group, use the candidate combination as the new best offset group; use the preferred mutation as the new initial distortion offset group, repeat until the distortion energy of the best offset group converges, and output the best offset group at this time as the distortion displacement field;

[0052] Perform pixel transfer on the distorted pixels based on the distortion displacement field, calculate the coordinates of the distorted pixels using the horizontal distortion offset and vertical distortion offset in the distortion displacement field (the horizontal axis value is summed with the horizontal distortion offset, and the vertical axis value is summed with the vertical distortion offset) to obtain the offset coordinates, replace the pixels at the offset coordinates with the distorted pixels, use the pixels directly adjacent to the distorted pixels as neighborhood pixels, and fill the pixel value at the distorted pixel with the pixel mean of the neighborhood pixels to obtain the oral cavity correction image; Filling based on the pixel mean can better preserve the details of the image and avoid over-smoothing of the image, ensuring that the quality of the corrected image reaches a relatively high level, especially having advantages in maintaining details and textures.

[0053] The methods for performing specular reflection compensation on the oral cavity correction image include:

[0054] Preset a set of scale components, perform light estimation on each pixel in the oral cavity correction image based on the set of scale components, and the formula for performing light estimation on each pixel in the oral cavity correction image is: where L(x, y) represents the light component of the pixel at pixel coordinates (x, y), CD represents the size of the set of scale components, w β represents the component weight of the β-th scale component in the set of scale components, used to control the influence of different scale components on the light component, and the component weight satisfies the component constraint condition, and the component constraint condition is: c β represents the β-th scale component in the set of scale components. F(c β , x, y) represents the convolution function, which is used to represent the convolution of the pixel at the pixel coordinates (x, y) with the kernel function based on the scale component c β . The commonly used kernel functions include one-dimensional Gaussian kernel function and two-dimensional Gaussian kernel function; the reflection evaluation is performed on each pixel in the oral cavity correction image based on the illumination component. The formula for performing the reflection evaluation on each pixel in the oral cavity correction image is:

[0055] where R(x, y) represents the reflection component of the pixel at the pixel coordinates (x, y), I(x, y) represents the pixel gray level of the pixel at the pixel coordinates (x, y), and τ represents the offset constant, which is used to smooth the reflection component; the adaptive compensation estimation is performed on each pixel in the oral cavity correction image based on the reflection component. The formula for performing the adaptive compensation estimation on each pixel in the oral cavity correction image is: where CH(x, y) represents the compensation factor of the pixel at the pixel coordinates (x, y), and γ represents the compensation control coefficient, which is used to adjust the size of the compensation factor and is set by those skilled in the art based on the actual situation. The larger the compensation control coefficient, the more obvious the compensation effect on the illumination change. When the compensation control coefficient is small, the image adjustment is smoother. represents the reflection gradient of the reflection component of the pixel at the pixel coordinates (x, y); the light compensation is performed on each pixel in the oral cavity correction image based on the compensation factor. The formula for performing the light compensation on each pixel in the oral cavity correction image is: IC(x, y) = I(x, y) 1+CH(x,y) ; where IC(x, y) represents the compensated pixel of the pixel at the pixel coordinates (x, y), and all the compensated pixels form the oral cavity compensation map;

[0056] The compensation of the oral cavity correction image based on the compensation factor keeps the edge and texture regions clear, avoids the loss of image details caused by over-smoothing, the shadow and reflection regions are compensated for brightness, and the lesion regions are made more obvious, facilitating subsequent lesion recognition; the set of scale components is the scale size for convolving the image. Common convolution scales include 3×3 scale, 5×5 scale, 7×7 scale, etc.; in this embodiment, the set of scale components is preferably 3×3 scale, 4×4 scale, and 5×5 scale.

[0057] The methods for lesion segmentation of the oral cavity compensation map include:

[0058] Preset pixel search radius. For pixel B in the oral cavity compensation map, the Euclidean distance formula is used to calculate the pixel distance between other pixels and pixel B, and other pixels with a coordinate distance less than the pixel search radius from pixel B form the pixel neighborhood set of pixel B; based on the pixel neighborhood, density evaluation is performed on each pixel in the oral cavity compensation map. The formula for gray density evaluation of each pixel in the oral cavity compensation map is: where MD a represents the pixel gray density of the ath pixel in the oral cavity compensation map, Size represents the size of the pixel neighborhood set, I a represents the pixel gray value of the ath pixel in the oral cavity compensation map, I j represents the pixel gray value of the jth pixel in the pixel neighborhood set, θ represents the gray control parameter used to control the gray difference, d aj represents the pixel distance between the ath pixel and the jth pixel in the pixel neighborhood set in the oral cavity compensation map, and r represents the pixel search radius; based on the pixel gray density, distance measurement is performed on the pixels in the oral cavity compensation map. The formula for distance measurement of the pixels in the oral cavity compensation map is: where Dis a represents the pixel distance of the ath pixel in the oral cavity compensation map, MD a represents the pixel gray density of the ath pixel in the oral cavity compensation map, MD j represents the pixel gray density of the jth pixel in the pixel neighborhood set, represents the density weight used to smooth the influence of density on the pixel distance, JG(MD a , MD j ) represents the filtering function, indicating that only pixels with MD j greater than MD a in the pixel neighborhood set participate in the calculation;

[0059] Based on the pixel gray density, use a clustering algorithm to cluster the pixels in the oral cavity compensation map to obtain pixel density clustering clusters; preset a density threshold, calculate the density mean of each pixel density clustering cluster, and mark the pixel density clustering clusters with a density mean greater than the density threshold as density lesion regions; Lesion regions usually appear as regions with relatively high local density, which may be formed by irregular tissue structures, masses, or lesions. These regions will be relatively dense, with fewer surrounding pixel points; Based on the pixel distance, use a clustering algorithm to cluster the pixels in the oral cavity compensation map to obtain pixel distance clustering clusters; preset a distance threshold, calculate the distance mean of each pixel density clustering cluster, and mark the pixel distance clustering clusters with a distance mean greater than the distance threshold as distance lesion regions; Lesion regions usually have a large distance in space from the normal region (non-lesion region) and a relatively large distance from the regions with lower density in other normal regions; Commonly used clustering algorithms include the K-Means clustering algorithm and the hierarchical clustering algorithm; Compare the regional overlap of the density lesion regions and the distance lesion regions, select the pixels that are both density lesion regions and distance lesion regions as lesion pixels, and crop the region where the lesion pixels in the oral cavity compensation map are aggregated to obtain a lesion sub-map.

[0060] The construction method of the oral health assessment model includes:

[0061] Based on the lesion sub-map and the corresponding image evaluation label, use the CNN model as the initial model of the oral health assessment model, use the lesion sub-map and the corresponding image evaluation label as the training data, use the training data as the training sample set, and use the training sample set to train the CNN model. Use the lesion sub-map and the corresponding image evaluation label as the input data of the oral health assessment model, and use the predicted evaluation label as the output data of the oral health assessment model; Use minimizing the error between the actual image evaluation label and the evaluation label predicted by the oral health assessment model as the training objective, use the recall rate function as the loss function of the oral health assessment model, and stop training to obtain the oral health assessment model when the loss function converges.

[0062] In this embodiment, by performing distortion conversion on oral image data, the distorted areas in oral images are effectively identified and corrected, ensuring the accuracy of the images and the retention of details. Especially when dealing with corners, curved surfaces, or reflective areas inside the oral cavity, the influence of deformation is effectively reduced, improving the image quality. By performing specular reflection compensation on the corrected oral images, the image deviation caused by uneven illumination can be eliminated. Especially under high-brightness light sources, the saliva surface has strong light reflection characteristics and is prone to generating strong specular reflection areas during imaging. The compensated images can better display the lesion areas, making the subsequent lesion recognition more accurate. By performing lesion segmentation on the oral compensation map, the lesion areas in the images are effectively identified, and the non-lesion areas are effectively filtered out, improving the accuracy of segmentation, precisely locating the lesion areas, and reducing human errors. Through the combination of big data analysis and deep learning models, the oral health of different patients can be accurately evaluated, thus promoting the realization of personalized oral care and health management.

[0063] Embodiment 2

[0064] Please refer to Figure 2 as shown. For the parts not described in detail in this embodiment, refer to the description content of Embodiment 1. A dental examination system based on big data analysis is provided, including:

[0065] Data acquisition and processing module: Collect oral image data and corresponding image evaluation labels, perform distortion conversion on the oral image data, and obtain corrected oral images;

[0066] Image compensation module: Perform specular reflection compensation on the corrected oral images to obtain oral compensation maps;

[0067] Region segmentation module: Perform lesion segmentation on the oral compensation maps to obtain lesion sub-images;

[0068] Model construction module: Construct an oral health assessment model based on the lesion sub-images and corresponding image evaluation labels, and achieve accurate assessment of oral health based on the oral health assessment model;

[0069] Each module is connected by wired and / or wireless means to achieve data transmission between modules.

[0070] Embodiment 3

[0071] This embodiment publicly provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it realizes the operation mode of the above-provided dental examination method based on big data analysis.

[0072] Since the electronic device introduced in this embodiment is the electronic device used to implement an oral examination method based on big data analysis in the embodiments of the present application, based on the oral examination method based on big data analysis introduced in the embodiments of the present application, those skilled in the art can understand the specific implementation manners and various variations of the electronic device in this embodiment. Therefore, the specific implementation of how this electronic device implements the method in the embodiments of the present application will not be introduced in detail here. As long as those skilled in the art implement the electronic device used in an oral examination method based on big data analysis in the embodiments of the present application, it falls within the scope of protection of the present application.

[0073] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain a formula that is closest to the actual situation. The preset parameters and threshold selection in the formulas are set by those skilled in the art according to the actual situation.

[0074] The above are only the preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for ordinary technical users in the technical field, several improvements and refinements made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.

Claims

1. An oral examination method based on big data analysis, characterized in that: include: S1, collecting oral image data and corresponding image evaluation labels, performing distortion transformation on the oral image data, and obtaining an oral correction image; S2, performing reflection compensation on the oral correction image to obtain an oral compensation image; S3, performing lesion segmentation on the oral compensation image to obtain a lesion sub-image; S4. Build an oral health assessment model based on the lesion sub-image and the corresponding image assessment labels, and achieve accurate assessment of oral health based on the oral health assessment model.

2. The oral examination method based on big data analysis according to claim 1, characterized in that: The method for acquiring the oral image data includes: using a high-definition oral endoscope equipped with a multi-angle imaging function and a macro lens, setting the angle and position of the device, combining a flexible arm mechanical assistance system, and utilizing the geometric alignment technology of multi-view images to splice images of different viewpoints, and the generated panoramic oral image is the oral image data.

3. The oral examination method based on big data analysis according to claim 2, characterized in that: The method of performing distortion conversion on the oral image data includes: The RGB weighted average method is used to grayscale the oral image data to obtain an oral grayscale image; for each pixel in the oral grayscale image, the pixel in the lower left corner of the oral grayscale image is used as the pixel center, and the pixel height and pixel width of the oral grayscale image are used as the coordinate scale to assign coordinates to each pixel to obtain the pixel coordinates of each pixel; the horizontal gradient operator in the edge detection operator is used to calculate the horizontal gradient of the pixels in the oral grayscale image to obtain the horizontal pixel gradient, and the vertical gradient operator in the edge detection operator is used to calculate the vertical gradient of the pixels in the oral grayscale image to obtain the vertical pixel gradient; the pixel tensor of each pixel is constructed based on the horizontal pixel gradient and the vertical pixel gradient, and the distortion of each pixel is evaluated based on the pixel tensor to obtain the distortion estimation; A distortion threshold is preset, and pixels whose distortion estimate is greater than or equal to the distortion threshold are marked as distorted pixels; a distortion displacement is explored for each pixel marked as a distorted pixel to obtain a distortion displacement field; pixel transfer is performed on the distorted pixels based on the distortion displacement field, and the coordinates of the distorted pixels are calculated using the horizontal distortion offset and the vertical distortion offset in the distortion displacement field to obtain offset coordinates, and the pixels at the offset coordinates are replaced by the distorted pixels, and the pixels directly adjacent to the distorted pixels are used as neighborhood pixels. The pixel values ​​at the distorted pixels are filled with the pixel mean of the neighborhood pixels to obtain a oral correction image.

4. The oral examination method based on big data analysis according to claim 3, characterized in that: The formula for evaluating the distortion of each pixel is: Among them, AS(x,y) represents the distortion estimate of the pixel at the pixel coordinate (x,y), |ZL(x,y)| represents the modulus of the pixel tensor at the pixel coordinate (x,y), Gx represents the horizontal pixel gradient, Gy represents the vertical pixel gradient, x represents the horizontal axis value of the pixel coordinate, and y represents the vertical axis value of the pixel coordinate.

5. The oral examination method based on big data analysis according to claim 4, characterized in that: The method of performing distortion displacement exploration on each pixel marked as a distorted pixel includes: Initialize the distortion offset group A, initialize the optimal offset group to a null value, the distortion offset group includes the horizontal distortion offset and the vertical distortion offset, and construct the distortion energy field function based on the distortion offset group. The expression of the distortion energy field function is: Among them, E represents the distortion energy, α1 represents the horizontal weight coefficient, represents the horizontal offset gradient, α2 represents the vertical weight coefficient, represents the vertical offset gradient, β represents the pixel distortion parameter, I(x,y) represents the pixel grayscale of the pixel at the pixel coordinate (x,y), I(x+u,y+v) represents the pixel grayscale of the pixel at the pixel coordinate (x+u,y+v), u represents the horizontal distortion offset, v represents the vertical distortion offset, dx represents the integration of the horizontal axis value, and dy represents the integration of the vertical axis value; the distortion energy of each distortion offset group is calculated based on the distortion energy field function, and the distortion offset group is initially selected using the tournament selection algorithm based on the distortion energy to obtain the initial distortion offset group; The initial distortion offset groups are combined in pairs to obtain offset pairs, and the horizontal distortion offset and the vertical distortion offset in each offset pair are recombined to obtain a child combination different from the initial distortion offset group; the distortion energy of each child combination is calculated by the distortion energy field function, and the child combination whose distortion energy is less than the distortion energy of the offset pair is taken as the new generation; a mutation scale is preset, and the mutation scale is an integer greater than zero. Based on the mutation scale, the horizontal distortion offset or the vertical distortion offset of the new generation is randomly mutated, and the horizontal distortion offset is randomly increased or decreased by one mutation scale, and the vertical distortion offset is randomly increased or decreased by one mutation scale. The scale is changed to obtain the mutation combination; the distortion energy of each mutation combination is calculated by the distortion energy field function, and the mutation combination with a distortion energy less than the distortion energy of the new generation is taken as the preferred mutation, and the preferred mutation with the smallest distortion energy is selected as the candidate combination. When the optimal offset group is a null value, the candidate combination is used as the new optimal offset group. When the optimal offset group is not a null value, if the distortion energy of the candidate combination is less than the distortion energy of the optimal offset group, the candidate combination is used as the new optimal offset group; the preferred mutation is used as the new initial distortion offset group, and the process is repeated until the distortion energy of the optimal offset group converges, and the optimal offset group at this time is output as the distortion displacement field.

6. The oral examination method based on big data analysis according to claim 5, characterized in that: The method of performing reflection compensation on the oral correction image includes: A set of scale components is preset, and illumination estimation is performed on each pixel in the oral correction image based on the set of scale components. The formula for illumination estimation for each pixel in the oral correction image is: Among them, L(x,y) represents the illumination component of the pixel at the pixel coordinate (x,y), CD represents the size of the scale component set, and w β Represents the component weight of the βth scale component in the scale component set. The component weight satisfies the component constraint condition. The component constraint condition is: c β represents the βth scale component in the scale component set, F(c β ,x,y) represents the convolution function; based on the illumination component, the reflection of each pixel in the oral correction image is evaluated to obtain the reflection component; based on the reflection component, the adaptive compensation estimation is performed on each pixel in the oral correction image to obtain the compensation factor; based on the compensation factor, light compensation is performed on each pixel in the oral correction image to obtain the compensated pixel, and all compensated pixels constitute the oral compensation map.

7. The oral examination method based on big data analysis according to claim 6, characterized in that: The formula for evaluating the reflection of each pixel in the oral correction image is: Where R(x,y) represents the reflection component of the pixel at the pixel coordinate (x,y), I(x,y) represents the pixel grayscale of the pixel at the pixel coordinate (x,y), and τ represents the offset constant; The formula for adaptive compensation estimation for each pixel in the oral correction image is: Among them, CH(x,y) represents the compensation factor of the pixel at the pixel coordinate (x,y), γ represents the compensation control coefficient, Represents the reflectance gradient of the reflectance component of the pixel at pixel coordinates (x, y).

8. The oral examination method based on big data analysis according to claim 7, characterized in that: The method of performing lesion segmentation on the oral compensation image includes: A pixel search radius is preset, and for pixel B in the oral compensation map, the pixel distance between other pixels and pixel B is calculated using the Euclidean distance formula, and other pixels whose coordinate distance to pixel B is less than the pixel search radius constitute a pixel neighborhood set of pixel B; density evaluation is performed on each pixel in the oral compensation map based on the pixel neighborhood to obtain pixel grayscale density; distance measurement is performed on pixels in the oral compensation map based on the pixel grayscale density to obtain pixel distance; Based on the pixel grayscale density, a clustering algorithm is used to cluster the pixels in the oral compensation image to obtain pixel density clustering clusters; a density threshold is preset, the density mean of each pixel density clustering cluster is calculated, and the pixel density clustering clusters with a density mean greater than the density threshold are marked as density lesion areas; based on pixel distance, a clustering algorithm is used to cluster the pixels in the oral compensation image to obtain pixel distance clustering clusters; a distance threshold is preset, the distance mean of each pixel density clustering cluster is calculated, and the pixel distance clustering clusters with a distance mean greater than the distance threshold are marked as distance lesion areas; a regional overlap comparison is performed on the density lesion area and the distance lesion area, and pixels that are both density lesion areas and distance lesion areas are screened out as lesion pixels, and the area where the lesion pixels in the oral compensation image are clustered is cropped to obtain a lesion sub-image.

9. The oral examination method based on big data analysis according to claim 8, characterized in that: The formula for evaluating the grayscale density of each pixel in the oral compensation image is: Among them, MD a represents the pixel grayscale density of the ath pixel in the oral compensation map, Size represents the size of the pixel neighborhood set, and I a represents the pixel grayscale of the ath pixel in the oral compensation image, I j represents the pixel grayscale of the jth pixel in the pixel neighborhood set, θ represents the grayscale control parameter, and d aj represents the pixel distance between the ath pixel in the oral compensation map and the jth pixel in the pixel neighborhood set, and r represents the pixel search radius; The formula for measuring the distance of pixels in the oral compensation map is: Among them, Dis a Represents the pixel distance of the ath pixel in the oral compensation map, MD a Represents the pixel grayscale density of the ath pixel in the oral compensation image, MD j represents the pixel grayscale density of the jth pixel in the pixel neighborhood set, represents density weight, JG(MD a ,MD j ) represents a filter function.

10. An oral examination system based on big data analysis, used to implement an oral examination method based on big data analysis as claimed in any one of claims 1 to 9, characterized in that: include: Data acquisition and processing module: collects oral image data and corresponding image evaluation labels, performs distortion conversion on the oral image data, and obtains oral correction images; Image compensation module: perform reflection compensation on the oral correction image to obtain the oral compensation image; Region segmentation module: performs lesion segmentation on the oral compensation image to obtain lesion sub-images; Model building module: Build an oral health assessment model based on the lesion sub-graph and the corresponding image assessment labels, and achieve accurate assessment of oral health based on the oral health assessment model.

Images