Caries identification method and device based on oral panoramic radiographs

Abstract

The present application discloses a method for identifying caries based on an oral panoramic film image and related equipment. The method includes obtaining an oral panoramic film image, and performing a preprocessing operation on the oral panoramic film image to obtain a target oral panoramic film image, wherein the preprocessing operation includes a denoising operation and / or a brightness adjustment operation; inputting the target oral panoramic film image into a trained image segmentation network model, and determining the tooth area in the oral panoramic film image through the image segmentation network model; inputting the tooth area into a trained caries detection network model, and determining the caries in the oral panoramic film image through the caries detection network model. The present application improves the accuracy and efficiency of caries identification by preprocessing an oral panoramic film image to obtain a target image, then performing image segmentation on the target image to obtain an oral region, and finally determining the caries in the target image through a detection network model.

Description

Technical Field

[0001] The present application relates to the field of biological technology, and in particular to a method and related device for identifying caries based on oral panoramic radiographs. Background Art

[0002] Early identification of oral health problems, especially caries, is crucial to preventing further tooth damage and providing timely treatment. Traditional caries identification methods usually rely on the subjective judgment of dentists and are easily influenced by subjective opinions and experience, leading to misdiagnosis or missed diagnosis. Traditional caries detection methods require professional dental expertise, which limits the popularity and application of detection methods. And because of the slow speed of manual interpretation and analysis, traditional methods are not suitable for large-scale oral health screening.

[0003] Therefore the prior art still needs to be improved and enhanced. Summary of the invention

[0004] The technical problem to be solved by the present application is to provide a method and related device for identifying caries based on oral panoramic radiographs in view of the deficiencies in the prior art.

[0005] In order to solve the above technical problems, the first aspect of the embodiment of the present application provides a method for identifying caries based on an oral panoramic radiograph image, the method comprising:

[0006] Acquire an oral panoramic film image, and perform a preprocessing operation on the oral panoramic film image to obtain a target oral panoramic film image, wherein the preprocessing operation includes a denoising operation and / or a brightness adjustment operation;

[0007] Inputting the target oral panoramic radiograph image into a trained image segmentation network model, and determining the tooth region in the oral panoramic radiograph image by means of the image segmentation network model;

[0008] The tooth region is input into a trained caries detection network model, and caries in the oral panoramic image are determined by the caries detection network model.

[0009] According to the above-mentioned technical means, the present application makes the oral panoramic film clearer through the pre-processing operation of denoising and / or brightness adjustment, providing a better target image for subsequent analysis. Secondly, the target image is input into the trained image segmentation network model to determine the tooth area in the oral panoramic film image. Compared with the traditional manual recognition method, this image recognition method using deep learning technology is not only more accurate, but also improves work efficiency. Finally, by inputting the determined tooth area into the trained caries detection network model, caries in the oral panoramic film image can be accurately identified. This detection method based on deep learning technology improves the efficiency and accuracy of caries detection.

[0010] In one implementation, the method for identifying caries based on oral panoramic radiographs, wherein the denoising operation specifically includes:

[0011] For each pixel in the oral panoramic film image, determining a denoising window corresponding to the pixel;

[0012] Determine the pixel median corresponding to the denoising window, and calculate the pixel weight based on the pixel median to obtain the weight corresponding to each pixel;

[0013] The oral panoramic film image is denoised based on the weight of each pixel.

[0014] According to the above technical means, the noise in the image can be eliminated or reduced, thereby improving the image quality and readability. When identifying caries, the noise in the image makes the process of identifying caries more difficult. The optimized image can accurately extract the image information, especially the information of the tooth part, through the calculated weights, so that medical staff can more accurately identify and determine caries in the mouth.

[0015] The above technical means obtain the weight of each pixel through the denoising window and pixel median calculation, making the denoising operation more accurate and personalized. This denoising method assigns weight to each pixel, processes the noise in the image more accurately, and improves the overall quality and clarity of the image.

[0016] In one implementation, the denoising window calculates a pixel median, wherein calculating the weight of the pixel based on the pixel median specifically includes:

[0017] Obtaining the maximum and minimum values ​​of the pixels in the denoising window, and removing the pixels corresponding to the maximum and minimum values ​​from the denoising window to obtain a candidate pixel set corresponding to the denoising window;

[0018] The pixel median corresponding to the denoising window is determined according to the pixel values ​​of all pixels in the candidate pixel set.

[0019] According to the above technical means, it helps to reduce or eliminate image noise and enhance the important features of the image. By removing the maximum and minimum pixel values, possible noise interference is eliminated, which usually manifests as abnormally high or low pixel values. These two types of pixels usually have a negative impact on the overall quality of the image. By removing them from the candidate pixel set, the impact of noise on the pixel weight calculation can be reduced, thereby obtaining more accurate pixel weights. Calculating the median pixel value can provide a more accurate representation of the pixel value than the simple average pixel value, because the median value will not be affected by outliers or noise pixels. In this way, when performing weight calculations, the results obtained will be more accurate and robust, thereby improving the ability to emphasize and retain important features in the image, thereby improving the clarity and quality of the image.

[0020] In one implementation, the method for identifying caries based on oral panoramic radiographs, wherein the brightness adjustment operation specifically includes:

[0021] Calculating the grayscale probability distribution of the oral panoramic film image;

[0022] Calculating a brightness adjustment parameter based on the grayscale probability distribution;

[0023] The brightness of the oral panoramic film image is adjusted based on the brightness adjustment parameter to obtain the oral panoramic film image after brightness adjustment.

[0024] According to the above technical means, the accuracy and efficiency of caries identification are improved. By calculating the grayscale probability distribution, the light and dark changes and image quality in the image are more accurately reflected, which helps to adjust the brightness of the image more accurately and can adapt to various types of oral panoramic images. Brightness adjustment can optimize image quality, make the area with caries more obvious, and facilitate accurate identification.

[0025] In one implementation, the method for identifying caries based on oral panoramic radiographs is characterized in that the target oral panoramic radiograph image is input into a trained image segmentation network model, and the tooth area is determined by the image segmentation network model, and the objective function of the image segmentation network model is:

[0026]

[0027] in, Represents the number of pixels in the oral panoramic image, and denotes the probability that the i-th pixel belongs to the tooth area and the non-tooth area, j denotes the weight, and d i,tooth and d i,nontooth denote the distance between the i-th pixel and the pixel center of the tooth area and the pixel center of the non-tooth area, respectively.

[0028] In one implementation, the objective function used in the training process of the caries detection network model is:

[0029]

[0030] Among them, ψ represents the number of grids that need to be identified by the caries detection network model, Indicates whether there is caries in the current detection grid. k and They represent the kth parameter value of the caries location identified by the caries detection network model and the actual location of the caries, respectively. The serial number b' indicates whether the current detection grid contains the real caries type, and α1 and α2 both represent control weights.

[0031] A second aspect of an embodiment of the present application provides a dental caries identification device based on an oral panoramic radiograph image, the device comprising:

[0032] A preprocessing module, used for acquiring an oral panoramic film image, and performing a preprocessing operation on the oral panoramic film image to obtain a target oral panoramic film image, wherein the preprocessing operation includes a denoising operation and / or a brightness adjustment operation;

[0033] A region segmentation module, used for inputting a target oral panoramic film image into a trained image segmentation network model, and determining a tooth region in the oral panoramic film image through the image segmentation network model;

[0034] The caries detection module is used to input the tooth area into a trained caries detection network model, and determine the caries in the oral panoramic image through the caries detection network model.

[0035] A third aspect of an embodiment of the present application provides a computer-readable storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the steps in any of the above-described methods for identifying caries based on oral panoramic radiographs.

[0036] A fourth aspect of the embodiments of the present application provides a terminal device, which includes: a processor and a memory;

[0037] The memory stores a computer-readable program executable by the processor;

[0038] When the processor executes the computer-readable program, the processor implements the steps in any of the above-mentioned methods for identifying caries based on oral panoramic radiographs.

[0039] Beneficial effects: Compared with the prior art, the present application provides a method for identifying caries based on oral panoramic film images and a related device, the method comprising acquiring an oral panoramic film image, and performing a preprocessing operation on the oral panoramic film image to obtain a target oral panoramic film image, wherein the preprocessing operation comprises a denoising operation and / or a brightness adjustment operation; inputting the target oral panoramic film image into a trained image segmentation network model, and determining the tooth region in the oral panoramic film image through the image segmentation network model; inputting the tooth region into a trained caries detection network model, and determining the caries in the oral panoramic film image through the caries detection network model. The present application obtains a high-quality target image by performing denoising and / or brightness adjustment preprocessing on the oral panoramic film image, and then performs image segmentation on the target image to obtain the oral region, optimizes the image quality, and makes the caries region more obvious, and finally determines the caries in the target image through the detection network model, thereby improving the accuracy and efficiency of caries identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without inventive work.

[0041] Figure 1 A flow chart of a caries identification method based on oral panoramic radiographs provided in this application.

[0042] Figure 2 This is a schematic diagram of the structure of the caries identification device based on oral panoramic radiographs provided in this application.

[0043] Figure 3 This is a schematic diagram of the structure of the terminal device provided in this application. DETAILED DESCRIPTION

[0044] The present application provides a method and a related device for identifying caries based on oral panoramic radiographs. To make the purpose, technical solution and effect of the present application clearer and more specific, the present application is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0045] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be an intermediate element. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0046] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with those in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.

[0047] It should be understood that the sequence numbers and sizes of the steps in this embodiment do not mean the order of execution. The execution order of each process is determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0048] Research has found that early identification of oral health problems, especially caries, is crucial to preventing further tooth damage and providing timely treatment. Traditional caries identification methods usually rely on the subjective judgment of dentists and are easily influenced by subjective opinions and experience, leading to misdiagnosis or missed diagnosis. Traditional caries detection methods require professional dental expertise, which limits the popularity and application of detection methods. And because of the slow speed of manual interpretation and analysis, traditional methods are not suitable for large-scale oral health screening.

[0049] In order to solve the above problems, in an embodiment of the present application, an oral panoramic film image is obtained, and a preprocessing operation is performed on the oral panoramic film image to obtain a target oral panoramic film image, wherein the preprocessing operation includes a denoising operation and / or a brightness adjustment operation; the target oral panoramic film image is input into a trained image segmentation network model, and the tooth area in the oral panoramic film image is determined by the image segmentation network model; the tooth area is input into a trained caries detection network model, and the caries in the oral panoramic film image is determined by the caries detection network model. The present application improves the accuracy and efficiency of caries recognition by preprocessing the oral panoramic film image to obtain a target image, then performing image segmentation on the target image to obtain an oral region, and finally determining the caries in the target image by detecting the network model.

[0050] The application content is further explained below through the description of embodiments in conjunction with the accompanying drawings.

[0051] This embodiment provides a method for identifying caries based on oral panoramic radiographs. Figure 1 As shown, the method includes:

[0052] S10, acquiring an oral panoramic film image, and performing a preprocessing operation on the oral panoramic film image to obtain a target oral panoramic film image, wherein the preprocessing operation includes a denoising operation and / or a brightness adjustment operation.

[0053] Specifically, oral panoramic radiographs, also known as full-mouth dental radiographs or oral panoramic X-rays, are a form of medical imaging commonly used in stomatology to show a comprehensive view of the teeth, jawbone, nasal cavity and other adjacent structures. The advantage of oral panoramic radiographs is that they can fully and accurately reveal the oral structure in one image, providing comprehensive reference information for clinicians.

[0054] Oral panoramic radiographs are saved as high-definition digital image files, such as JPEG, PNG, or TIFF formats, where the number of bits per pixel is usually 8 or 16, which can reflect a sufficient grayscale range to reveal the subtle structures in the oral cavity. Like other medical imaging images, the size (resolution) of oral panoramic radiographs depends on specific diagnostic needs and is also affected by the settings of the acquisition equipment. Typically, for human oral structures, the image size may reach 2000×2000 pixels or even higher to ensure sufficient clarity.

[0055] After obtaining the oral panoramic radiograph, preprocessing is performed to improve its quality and ensure that the subsequent steps can proceed smoothly. Common preprocessing operations include denoising and brightness adjustment. Denoising can use median filtering, Gaussian filtering, etc. to eliminate various noises in the image caused by the shooting equipment or environment. Brightness adjustment adjusts the contrast and brightness of the image to compensate for the problem of too dark or too bright images caused by insufficient or excessive light during shooting.

[0056] The quality of oral panoramic film images is standardized through preprocessing operations, which provides a basis for subsequent more in-depth image processing and analysis. The preprocessing operation can flexibly select denoising, brightness adjustment, or both as needed. After the above preprocessing operations, the final oral panoramic film image is the target oral panoramic film image, which not only retains all the important information in the original image, but also has higher usability.

[0057] In one implementation, the denoising operation specifically includes:

[0058] S11, for each pixel in the oral panoramic image, determining a denoising window corresponding to the pixel;

[0059] S12, determining the pixel median corresponding to the denoising window, and calculating the pixel weight based on the pixel median to obtain the weight corresponding to each pixel;

[0060] S13, denoising the oral panoramic image based on the weight of each pixel.

[0061] Specifically, in step S11, the target oral panoramic image contains the complete oral structure information of the individual. Each pixel of the target oral panoramic image has a unique position coordinate and a corresponding grayscale value, and the information it carries is closely related to its spatial position and adjacent pixels in the oral panoramic image. Therefore, this step processes each pixel in the oral panoramic image, emphasizing the uniqueness of each pixel and its relationship with surrounding pixels.

[0062] The denoising window is a square or rectangular area of ​​a preset size, and the pixel is located at the center of the window. From a technical point of view, the size of the window may be adjusted according to the specific denoising needs and the unique situation of pixel distribution. For example, in some complex or dense areas, a smaller window is required to eliminate noise more finely; while in some relatively simple or loose areas, a larger window is required to achieve more effective denoising. In the specific implementation of the present application, the denoising window is determined with each pixel in the target oral panoramic film image as the center, and the denoising window size can be adjusted according to the specific application, such as 3x3, 5x5, etc.

[0063] In step S12, each denoising window covers the area around the target pixel, and the grayscale values ​​of each pixel in this area constitute a data set, so that the pixel median corresponding to the denoising window can be determined by determining the median of the data set, wherein the pixel median reflects the average brightness condition of the area around the target pixel. In addition, the weight application reflects the importance of each pixel in the target oral panoramic image. There are many ways to determine the weight according to the pixel median determined by the denoising window corresponding to the pixel. For example, the value of each pixel is compared with its corresponding pixel median. If the pixel value is far from the median, that is, it may be noise, the weight of this pixel will be relatively low; otherwise, the weight is high, so that the influence of noise pixels on the final image quality can be controlled by the median.

[0064] In one implementation of the embodiment of the present application, the weight determination process may adopt a weight allocation method. Specifically, each pixel in the denoising window is considered one by one, and compared with the median pixel value of the window. If the value of the pixel is close to the median, the pixel is considered to be relatively reliable and is assigned a high weight; conversely, if the value of the pixel is far from the median, it is considered to be a possible noise and is assigned a low weight.

[0065] In this way, the weight value corresponding to each pixel is calculated. This weight value represents the importance of the pixel in the image denoising process, thus providing useful information for the subsequent image processing steps. By using this weight-based processing method, the quality of oral panoramic images is further improved, making the diagnostic information obtained from these images more accurate and detailed.

[0066] In an implementation of the embodiment of the present application, determining the pixel median corresponding to the denoising window specifically includes:

[0067] S121, obtaining the maximum value and the minimum value of the pixels in the denoising window, and removing the pixels corresponding to the maximum value and the minimum value from the denoising window to obtain a candidate pixel set corresponding to the denoising window;

[0068] S122. Determine a pixel median corresponding to the denoising window according to the pixel values ​​of all pixels in the candidate pixel set.

[0069] Specifically, in steps S121 and S122, the specific implementation of the present application is to remove the maximum and minimum values ​​in the denoising window obtained in step S11, obtain the median τ of the remaining pixels, and obtain the calculation formula of the median τ as follows:

[0070] τ=Median(pixels)

[0071] Among them, pixels are the pixels remaining after the denoising window removes the maximum and minimum values ​​in the window, that is, the candidate pixel set; the Median function is used to obtain the median.

[0072] Calculating the weight of the pixel based on the pixel median to obtain the weight corresponding to each pixel in the specific implementation of the present application includes:

[0073] The weight of each pixel is calculated as follows:

[0074]

[0075] Wherein, N represents the number of pixels in pixels; n=1,2,…,N; n'=1,2,…,N; The calculation method is:

[0076]

[0077] Among them, Δ n is the absolute difference, which is calculated as:

[0078] Δ n =|pixels n -τ|

[0079] Among them, pixels n It represents the nth pixel among the pixels remaining after the denoising window removes the maximum and minimum values ​​in the window; is the mean of the absolute differences, calculated as:

[0080]

[0081] In step S13, the specific process of denoising the oral panoramic film image based on the weights corresponding to the pixels obtained in step S12 includes: traversing all pixels of the oral panoramic film image, and calculating new pixel values ​​according to the weights of the pixels and the pixel values ​​in the denoising window where they are located. Generally, the higher the weight of the pixel in the window, the greater the influence of its pixel value on the new pixel value. In other words, pixels with high weights have a greater decisive influence on the denoising effect.

[0082] For example, a simple weighted average denoising algorithm is used, where the weight of each pixel is proportional to its grayscale value. In this case, all pixels are traversed and all pixels in the denoising window of each pixel are multiplied by their weights, then summed, and finally the sum is divided by the sum of the weights to obtain a new pixel value. The above weighted average denoising algorithm is beneficial for preserving the edge information of the image, because the edge is usually composed of pixels with higher grayscale values.

[0083] The pixel values ​​of the original oral panoramic film image are replaced with new pixel values, and the new oral panoramic film image obtained is the denoised image.

[0084] Through the weight-based denoising method, the complex information in oral panoramic radiographs can be effectively utilized to improve the quality of denoised images, thereby further improving the accuracy and efficiency of dental diagnosis.

[0085] In the specific implementation of the present application, the denoised pixel is obtained according to the calculated weight, and the formula is:

[0086]

[0087] Among them, w n' is the pixel weight, and ω is the oral panoramic image obtained after weighted denoising.

[0088] In summary, oral panoramic radiographs may be affected by negative factors such as noise and artifacts, resulting in poor image quality. Denoising can effectively reduce the noise in the image, making the oral structure clearer and more recognizable, thereby improving the image quality. The denoised image is easier to be recognized and analyzed by subsequent processing algorithms. In the caries recognition method, if the image quality is poor, it may affect the extraction and analysis of features, and thus affect the accuracy of recognition.

[0089] In one implementation, the brightness adjustment operation specifically includes:

[0090] S101, calculating the grayscale probability distribution of the oral panoramic image;

[0091] S102, calculating a brightness adjustment parameter based on the grayscale probability distribution;

[0092] S103 . Adjust the brightness of the oral panoramic film image based on the brightness adjustment parameter to obtain the oral panoramic film image after brightness adjustment.

[0093] Specifically, the brightness adjustment operation usually includes increasing or decreasing the brightness value of the oral panoramic image. Generally speaking, the selection of the overall brightness of the image mainly depends on the initial quality of the image, and the basic brightness of the image does not change the authenticity of the internal structure of the oral cavity presented. While ensuring the image quality, the brightness is increased as much as possible to correct the image. The image brightness is adjusted automatically or manually by using image preprocessing software. During the brightness adjustment process, the contrast can also be adjusted when necessary. Improving the image contrast can increase the difference between two or more colors, making the details clearer and easier to identify and analyze various structures and diseases in the oral cavity. After the brightness is adjusted to a suitable level, preview whether the modified image meets the requirements. The image that meets the requirements is the target oral panoramic film image obtained. This panoramic oral image with brightness adjustment will provide a clearer and easier to interpret visual resource for the subsequent oral diagnosis and analysis.

[0094] Further, in step S101, the target oral panoramic film image is converted into a grayscale image. This means that the original color image is converted into an image containing only grayscale information. In this process, the color information will be lost, but important information such as structure and shape will still be retained. The main purpose of converting the image into a grayscale image is to improve the efficiency of subsequent processing, simplify the computational complexity, and make the grayscale probability distribution calculation in the subsequent steps easier and faster.

[0095] The grayscale probability distribution of this grayscale image is calculated by counting the frequency of each grayscale level in the image. In the field of image processing or analysis, the grayscale probability distribution counts all the pixel values ​​of the image, and the result is a chart or function representing all possible grayscale levels and their corresponding frequencies in the image. For a given image, its grayscale probability distribution provides rich information, which can reveal the image's contrast, brightness, structure and other characteristics.

[0096] In the specific implementation of the present application, the grayscale probability distribution of the denoised oral panoramic film image is first calculated, and then truncation is performed based on the grayscale probability distribution. The specific formula for calculating the grayscale probability distribution of the denoised oral panoramic film image is:

[0097]

[0098] Among them, p l represents the probability of pixels with gray value l in the denoised oral panoramic image, l = 0, 1, 2, ..., 255; count(l) represents the number of pixels with gray value l; Indicates the number of pixels in the denoised oral panoramic film.

[0099] The grayscale probability distribution is truncated, and the truncation method is:

[0100]

[0101] Among them, δ is the cutoff threshold, which is calculated as:

[0102]

[0103] in, express The maximum value of

[0104] In step S102, based on the grayscale probability distribution obtained in step S101, a brightness adjustment parameter is calculated. The so-called brightness adjustment parameter is mainly used to change a specific value of the image brightness. The brightness adjustment parameter can be the slope and intercept of a linear function, which is used to map each grayscale level of the image to a new level, and the mapped grayscale level is more in line with the target brightness requirement. In more complex situations, the brightness adjustment parameter can be a parameter of a nonlinear function to specifically adjust the brightness of a specific area.

[0105] In order to obtain the brightness adjustment parameters, it is necessary to meet the preset brightness requirements and the visual effects of the image. Ideally, it is necessary to find a moderate brightness adjustment parameter setting based on the standards and habits of human eye observation, but this method usually requires multiple attempts and the experience and judgment of image processing experts. In addition, some automated parameter optimization methods can also be used, such as gradient descent method or genetic algorithm, etc. The goal is to find a parameter that makes the image meet the preset requirements and has the best visual effect.

[0106] In a specific implementation of the present application, according to the truncated grayscale probability distribution, the formula for calculating the brightness adjustment parameter corresponding to each grayscale is:

[0107]

[0108] Among them, γ l represents the brightness adjustment parameter of the pixel with gray value l; ρ is the brightness adjustment parameter cutoff value, and in the specific implementation of this application, ρ=0.5.

[0109] In step S103, the brightness adjustment parameters obtained in step S102 are applied to the oral panoramic radiograph image. The operation of applying to the image is performed by a point-by-point operation, that is, each pixel finds a corresponding value in the brightness adjustment parameter according to its grayscale value, and adjusts its grayscale level. This step needs to be performed the same number of times as the number of pixels.

[0110] The oral panoramic image after brightness adjustment is easier to interpret than the denoised oral panoramic image because it has been adjusted to a form that better meets the preset brightness requirements. In some applications that rely on depth information, such as the diagnosis of dental diseases, the brightness-adjusted image can provide more useful information.

[0111] In the specific implementation of the present application, based on the brightness adjustment parameter adjustment The formula for the brightness of each pixel in is:

[0112]

[0113] in, Represents the pixel with gray value l in the denoised oral panoramic image; after completing the brightness adjustment, the preprocessed oral panoramic image is obtained

[0114] Brightness adjustment can make the oral structure in the oral panoramic film image more visible, highlight the details of the relevant area, and provide more visual information for subsequent feature extraction and analysis. Brightness adjustment can make the oral panoramic film image more consistent in terms of light and darkness, which helps to establish consistency standards between different oral panoramic film images, thereby improving the repeatability of caries identification methods. Oral panoramic film images with brightness adjustment can better highlight the characteristics of teeth, making oral problems such as caries lesions easier to identify and analyze, thereby improving the accuracy of identification.

[0115] S20, inputting the target oral panoramic radiograph image into a trained image segmentation network model, and determining the tooth region in the oral panoramic radiograph image by means of the image segmentation network model.

[0116] Specifically, the trained image segmentation network model is used to determine the tooth area in the input oral panoramic film image. The image segmentation network model has generalization ability and can process a variety of oral panoramic film images. According to the preset model structure and parameter settings, the input oral panoramic film image is segmented into the target area, that is, the tooth area.

[0117] Generally speaking, the process of segmenting an oral panoramic radiograph includes: inputting the oral panoramic radiograph image into an image segmentation network model. The image segmentation network model is based on some well-known deep learning architectures, such as U-Net, Mask R-CNN or other convolutional neural networks.

[0118] Inside the model, the oral panoramic radiograph is first processed through a series of convolutional layers, activation function layers, and pooling layers. These layers extract and learn key features in the image. For example, basic structures such as edges and colors are first identified through low-level feature extraction layers. Mid-level and high-level feature extraction layers identify more complex features such as tooth shape, size, and relative position.

[0119] At the end of the network, a softmax or sigmoid function is applied to calculate the probability of whether each image segmentation network model is a tooth area. According to the set threshold, the image segmentation network model is segmented to determine the tooth area in the oral panoramic image.

[0120] In one implementation, for a panoramic oral film image, assume that there are 32 teeth in the image. First, input this image into a pre-trained U-Net network model. Inside the model, the original input image is processed by a series of convolutional layers, activation function layers, and pooling layers to learn the key features in the image. After network processing, the probability of whether each pixel is a tooth area is mapped out. Set a threshold, such as 0.5, and mark the pixels with a probability higher than 0.5 as tooth areas to form a segmentation map. By analyzing the segmentation map, the 32 teeth in the panoramic oral film image can be identified and marked.

[0121] After the tooth area is determined, further analysis and processing are performed on these areas, such as tooth recognition, tooth arrangement, tooth health status analysis, etc. With the help of image segmentation technology, efficient and accurate tooth area positioning and further analysis of teeth can be achieved, which will improve the efficiency of the oral medical industry, provide doctors with more accurate diagnostic information, and provide patients with better medical services.

[0122] In the implementation of the present application, the objective function of the image segmentation network model is:

[0123]

[0124] in, Represents the number of pixels in the oral panoramic image, and denotes the probability that the i-th pixel belongs to the tooth area and the non-tooth area, j denotes the weight, and d i,tooth and d i,nontooth denote the distance between the i-th pixel and the pixel center of the tooth area and the pixel center of the non-tooth area, respectively.

[0125] d i,tooth and d i,nontooth The calculation formula is:

[0126] d i,tooth=(x i -c tooth ) 2

[0127] d i,nontooth =(x i -c nontooth ) 2

[0128] Among them, x i is the value of the ith pixel of the preprocessed oral panoramic image; c tooth and c nontooth They are divided into the pixel center of the tooth area and the pixel center of the non-tooth area.

[0129] Typically, the iterative solution process for optimizing segmentation of an image segmentation network model includes the following steps:

[0130] Update the probability of each pixel belonging to the tooth area and the non-tooth area, and calculate the probability of each pixel belonging to the tooth area based on the image segmentation network model. The probability is updated based on the current segmentation result and pixel features, such as color, texture, edge and other information, using a preset probability model (such as a Bayesian model, a Markov model, etc.).

[0131] The distance between each pixel and the pixel center of the tooth area and the pixel center of the non-tooth area is updated to update the spatial position relationship of each pixel relative to the segmentation result. The distance between the pixel centers can be calculated based on the Euclidean distance or other similarity measurement methods. The pixel centers of the tooth area and the non-tooth area can be determined by a certain clustering algorithm (such as the K-means algorithm).

[0132] The above two steps are repeated to continuously improve the accuracy of the segmentation results through the iterative optimization process. When the pixel center of the tooth area and the pixel center of the non-tooth area no longer change, that is, when the iteration reaches convergence, the segmentation process is considered complete. At this time, the obtained segmentation result clearly distinguishes the tooth area from other areas in the preprocessed oral panoramic image.

[0133] In the specific implementation of the present application, the formula for updating the probability of each pixel belonging to the tooth area and the non-tooth area is:

[0134]

[0135]

[0136] The formula for updating the distance between each pixel and the pixel center divided into the tooth area and the pixel center of the non-tooth area is:

[0137]

[0138]

[0139] S30, inputting the tooth region into a trained caries detection network model, and determining caries in the oral panoramic image through the caries detection network model.

[0140] Specifically, a tooth region separated from a panoramic oral radiograph is used as input, the tooth region may contain multiple teeth, and the goal of the detection network model is to find possible caries in these teeth.

[0141] Once the tooth region is determined, it is input into a trained caries detection network model. The caries detection network model is trained to learn how to obtain and analyze relevant features from the tooth region image, so as to accurately locate the location of caries.

[0142] The training process of the caries detection network model is carried out through a large number of oral panoramic film samples, including various tooth images with and without caries. During the training process, the network model will gradually learn and understand what kind of features and patterns may represent caries, and then it can be used to accurately identify and mark caries when new tooth area images are provided.

[0143] The caries locations marked by the network model will be provided to professional doctors or robots for further treatment. At the same time, the recognition results can also be used as an auxiliary tool for doctors to help more quickly and accurately assess the oral health status of patients and develop personalized treatment plans for patients.

[0144] In one implementation, the Faster R-CNN deep learning model is used to determine the location of caries in a panoramic oral radiograph.

[0145] First, the tooth area image is input into the neural network and features are extracted through deep learning of Faster R-CNN. The Faster R-CNN model is a type of convolutional neural network designed to quickly and accurately detect target objects in images, suitable for teeth and caries location needs.

[0146] The Faster R-CNN model mainly consists of two parts: one is the Region Proposal Network (RPN), which is used to generate candidate boxes; the other is the Fast R-CNN network, which is used to classify candidate boxes and fine-tune their positions.

[0147] In the process of detecting the target object in the image, Faster R-CNN first searches for the location where caries may exist in the tooth area image through the RPN network and generates a series of candidate boxes. Then, through the Fast R-CNN network, these candidate boxes are classified as caries or non-caries, and the position of the candidate boxes is fine-tuned to make them match the corresponding tooth area more accurately.

[0148] For example, in a panoramic oral image, there is a tooth in a carious state. After being processed by the Faster R-CNN model, the model can successfully locate the carious tooth and generate an accurate candidate box to surround it. At the same time, the model can also output a classification result indicating that it is a carious tooth.

[0149] After Faster R-CNN model training and recognition, the oral panoramic images clearly and accurately show the teeth with caries and their locations. This brings great convenience and accuracy to the subsequent dental treatment, thereby improving the level and efficiency of medical services.

[0150] In the implementation of the present application, the caries location and classification network is first defined as:

[0151]

[0152] DECAY is a caries detection network model based on Yolo target recognition network; type = {type1, type2, …, type H}, represents the probability of caries with different caries degrees identified by the caries detection network model, H represents the number of caries types. In this embodiment, the caries types that need to be classified are no caries, shallow caries, moderate caries and deep caries; loc = {loc1, loc2, loc3, loc4}, represents the location parameters of caries identified by the caries detection network model, including the length, width and center position of the caries outer bounding box; W is the parameter of the caries detection network model.

[0153] The objective function of the caries detection network model is:

[0154]

[0155] Wherein, h = 1, 2, …, ψ, ψ represents the number of grids that the caries detection network model needs to identify, which is obtained by evenly dividing the preprocessed oral panoramic film image; Indicates whether there is caries in the current detected grid. The judgment function is 1 if caries exists, otherwise it is 0; k = 1, 2, 3, 4; loc k and Respectively represent the kth parameter value of the caries location identified by the caries location and classification network and the actual caries location; b = 1, 2, ..., H; The serial number b' indicates whether the previously detected grid contains a real caries type, which is 1 if it contains a real caries type, otherwise it is 0; α1 and α2 represent control weights, and in the specific implementation of the present invention, α1=5 and α2=1.2 are preset.

[0156] In addition, in the specific implementation of the present invention, the caries detection network model is optimized based on gradient descent, and the optimized network is used to perform caries recognition on the patient's oral panoramic film image:

[0157] The caries detection network model is optimized based on gradient descent, and the gradient descent process is:

[0158]

[0159] Among them, W f and W f+1 They are the f-th and f+1-th optimization results of the parameters of the caries detection network model; the calculation method of Grad1 is:

[0160]

[0161] in, To optimize the target W f The partial derivative of θ represents the learning rate of the caries detection network model; Grad2 is calculated as follows:

[0162]

[0163]

[0164]

[0165] Wherein, f represents the number of optimizations; η1 and η2 are adjustment coefficients, and in the specific implementation of the present invention, η1=0.9, η2=0.99; ε=0.000001 are preset.

[0166] In summary, the present embodiment provides a method for dental caries recognition based on oral panoramic film images, the method comprising acquiring an oral panoramic film image, and performing a preprocessing operation on the oral panoramic film image to obtain a target oral panoramic film image, wherein the preprocessing operation comprises a denoising operation and / or a brightness adjustment operation; inputting the target oral panoramic film image into a trained image segmentation network model, and determining the tooth region in the oral panoramic film image through the image segmentation network model; inputting the tooth region into a trained caries detection network model, and determining the caries in the oral panoramic film image through the caries detection network model. The present application improves the accuracy and efficiency of dental caries recognition by preprocessing an oral panoramic film image to obtain a target image, then performing image segmentation on the target image to obtain an oral region, and finally determining the caries in the target image through a detection network model.

[0167] Based on the above-mentioned caries identification method based on oral panoramic radiographs, this embodiment provides a caries identification device based on oral panoramic radiographs, such as Figure 2 As shown, the device comprises:

[0168] A preprocessing module 100 is used to obtain an oral panoramic film image and perform a preprocessing operation on the oral panoramic film image to obtain a target oral panoramic film image, wherein the preprocessing operation includes a denoising operation and / or a brightness adjustment operation;

[0169] The region segmentation module 200 is used to input the target oral panoramic film image into a trained image segmentation network model, and determine the tooth region in the oral panoramic film image through the image segmentation network model;

[0170] The caries detection module 300 is used to input the tooth area into a trained caries detection network model, and determine caries in the oral panoramic image through the caries detection network model.

[0171] Based on the above-mentioned caries identification method based on oral panoramic radiograph images, this embodiment provides a computer-readable storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the steps in the caries identification method based on oral panoramic radiograph images as described in the above-mentioned embodiment.

[0172] Based on the above-mentioned caries identification method based on oral panoramic film images, the present application also provides a terminal device, such as Figure 3As shown, it includes at least one processor (processor) 20; display screen 21; and memory (memory) 22, and may also include a communications interface (Communications Interface) 23 and a bus 24. Among them, the processor 20, the display screen 21, the memory 22 and the communication interface 23 can communicate with each other through the bus 24. The display screen 21 is configured to display a preset user guide interface in the initial setting mode. The communication interface 23 can transmit information. The processor 20 can call the logic instructions in the memory 22 to execute the method in the above embodiment.

[0173] In addition, the logic instructions in the memory 22 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product.

[0174] The memory 22 is a computer-readable storage medium that can be configured to store software programs, computer executable programs, such as program instructions or modules corresponding to the methods in the embodiments of the present disclosure. The processor 20 executes functional applications and data processing by running the software programs, instructions or modules stored in the memory 22, that is, implementing the methods in the above embodiments.

[0175] The memory 22 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the terminal device, etc. In addition, the memory 22 may include a high-speed random access memory and may also include a non-volatile memory. For example, a variety of media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, may also be a transient storage medium.

[0176] In addition, the specific process of loading and executing the multiple instruction processors in the above-mentioned storage medium and the terminal device has been described in detail in the above-mentioned method, and will not be described one by one here.

[0177] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for identifying caries based on oral panoramic radiographs, characterized in that: The method comprises: Acquire an oral panoramic film image, and perform a preprocessing operation on the oral panoramic film image to obtain a target oral panoramic film image, wherein the preprocessing operation includes a denoising operation; Inputting the target oral panoramic radiograph image into a trained image segmentation network model, and determining the tooth region in the oral panoramic radiograph image by means of the image segmentation network model; Inputting the tooth region into a trained caries detection network model, and determining caries in the oral panoramic image through the caries detection network model; The denoising operation specifically includes: For each pixel in the oral panoramic film image, determining a denoising window corresponding to the pixel; Determine the pixel median corresponding to the denoising window, and calculate the pixel weight based on the pixel median to obtain the weight corresponding to each pixel; De-noising the oral panoramic film image based on the weight of each pixel; The step of calculating the weight of the pixel based on the pixel median to obtain the weight corresponding to each pixel specifically includes: Comparing each pixel within the denoising window with the pixel median; When the value of the pixel is close to the pixel median, a high weight is given to the pixel; When the value of the pixel is far away from the pixel median, a low weight is given to the pixel; The weight is calculated as follows: in, express The number of pixels in the pixel; ； ; The calculation method is: in, is the absolute difference, which is calculated as: in, It indicates the number of pixels remaining after the denoising window removes the maximum and minimum values ​​in the window. pixels; is the mean of the absolute differences, calculated as: in, The pixels remaining after removing the maximum and minimum values ​​in the denoising window are the candidate pixel set; Function is used to obtain the median value.

2. The method for identifying caries based on oral panoramic radiographs according to claim 1, characterized in that: Determining the pixel median corresponding to the denoising window specifically includes: Obtaining the maximum and minimum values ​​of the pixels in the denoising window, and removing the pixels corresponding to the maximum and minimum values ​​from the denoising window to obtain a candidate pixel set corresponding to the denoising window; The pixel median corresponding to the denoising window is determined according to the pixel values ​​of all pixels in the candidate pixel set.

3. The method for identifying caries based on oral panoramic radiographs according to claim 1, characterized in that: The brightness adjustment operation specifically includes: Calculating the grayscale probability distribution of the oral panoramic film image; Calculating a brightness adjustment parameter based on the grayscale probability distribution; The brightness of the oral panoramic film image is adjusted based on the brightness adjustment parameter to obtain the oral panoramic film image after brightness adjustment.

4. The method for identifying caries based on oral panoramic radiographs according to claim 1, characterized in that: The objective function used in the training process of the image segmentation network model is: in, Represents the number of pixels in the oral panoramic image, and Respectively represent The probability that a pixel belongs to the tooth area and the non-tooth area, represents the weight, and Respectively represent The distance between the pixel center of the tooth area and the pixel center of the non-tooth area.

5. The method for identifying caries based on oral panoramic radiographs according to claim 1, characterized in that: The objective function used in the training process of the caries detection network model is: in, represents the number of grids that need to be identified by the caries detection network model, Indicates the judgment function of whether there is caries in the current detection grid, and The first and second positions of the caries locations identified by the caries detection network model and the actual caries locations are respectively The value of the parameter, Indicates whether the current detection grid contains the serial number of the real caries type , and Both represent control weights, It indicates the probability of teeth with the same caries degree identified by the caries detection network model when the number of species is b.

6. A dental caries identification device based on oral panoramic radiographs, characterized in that: The device comprises: A preprocessing module, used for acquiring an oral panoramic film image, and performing a preprocessing operation on the oral panoramic film image to obtain a target oral panoramic film image, wherein the preprocessing operation includes a denoising operation; A region segmentation module, used for inputting a target oral panoramic film image into a trained image segmentation network model, and determining a tooth region in the oral panoramic film image through the image segmentation network model; A caries detection module, used for inputting the tooth region into a trained caries detection network model, and determining caries in the oral panoramic film image through the caries detection network model; The denoising operation specifically includes: For each pixel in the oral panoramic film image, determining a denoising window corresponding to the pixel; Determine the pixel median corresponding to the denoising window, and calculate the pixel weight based on the pixel median to obtain the weight corresponding to each pixel; De-noising the oral panoramic film image based on the weight of each pixel; The step of calculating the weight of the pixel based on the pixel median to obtain the weight corresponding to each pixel specifically includes: Comparing each pixel within the denoising window with the pixel median; When the value of the pixel is close to the pixel median, a high weight is given to the pixel; When the value of the pixel is far away from the pixel median, a low weight is given to the pixel; The weight is calculated as follows: in, express The number of pixels in the pixel; ； ; The calculation method is: in, is the absolute difference, which is calculated as: in, It indicates the number of pixels remaining after the denoising window removes the maximum and minimum values ​​in the window. pixels; is the mean of the absolute differences, calculated as: in, The pixels remaining after removing the maximum and minimum values ​​in the denoising window are the candidate pixel set; Function is used to obtain the median value.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in the caries identification method based on oral panoramic radiographs as described in any one of claims 1 to 5.

8. A terminal device, characterized in that: include: Processor and memory; The memory stores a computer-readable program executable by the processor; When the processor executes the computer-readable program, the steps of the caries identification method based on oral panoramic radiographs as described in any one of claims 1 to 5 are implemented.

Images