A Tooth Region Detection Method, Device, Equipment and Storage Medium
By adopting a semantic segmentation and object detection model with parallel processing in stomatological imaging processing and screening with the output results, the problem of inaccurate tooth area identification and segmentation in the prior art is solved, and higher recognition accuracy and segmentation clarity are achieved.
Patent Information
- Application Number
- CN202111320584.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-11-09
- Publication Date
- 2025-07-01
- Estimated Expiration
- 2041-11-09
AI Technical Summary
The prior art is difficult to quickly and accurately identify and segment the dental areas in stomatological images, resulting in a decrease in recognition efficiency and accuracy of doctors during the film reading process.
The semantic segmentation model and object detection model with parallel processing are integrated into a framework to identify and segment oral medical images. The screening module combines the output results of the two models to remove error areas to obtain accurate identification and clear segmentation of tooth areas.
It improves the accuracy of identification and segmentation clarity of teeth areas, reduces the human interference factors during the video reading process and the dependence on doctors' professional qualities, and further improves accuracy, consistency and credibility.
Smart Images
Figure CN114004970B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of oral medical imaging, and particularly to a method, device, equipment and storage medium for detecting tooth regions. Background Art
[0002] With the development and progress of society, people's living standards have been greatly improved, and their lifestyles have become increasingly diversified. This has also led to many oral problems caused by bad living habits, and people have paid more and more attention to the condition of their oral teeth.
[0003] According to oral images, the overall oral structure can be observed, and then case data can be recorded. Oral doctors first make a visual judgment on the teeth. They observe whether there are damages and color changes of the teeth under the light source through their eyes, and then detect the condition of the teeth through special oral examination instruments, and conduct a comprehensive analysis in combination with oral panoramic films. Although oral images have the characteristics of high universality and low cost, doctors still need to judge teeth, gums and other oral regions based on oral panoramic films. Since human film reading is affected by emotions and fatigue, long-term film reading will lead to a decrease in the recognition efficiency of doctors. At the same time, oral images are grayscale images, and it is difficult for the naked eye to distinguish regions with unclear target edge contours in grayscale images. Therefore, currently imaging the entire oral cavity cannot enable doctors to quickly identify teeth and other oral regions. Summary of the Invention
[0004] In view of this, the purpose of the present invention is to provide a method, device, equipment and storage medium for detecting tooth regions, which can effectively obtain a tooth region with accurate recognition and clear segmentation. The specific scheme is as follows:
[0005] A method for detecting tooth regions includes:
[0006] Obtain an oral medical image dataset; the oral medical image dataset includes oral medical images and oral medical image data annotated on the oral medical images;
[0007] Preprocess the oral medical image dataset, and use the preprocessed oral medical image dataset as a training sample;
[0008] Construct an oral tooth prediction model; the oral tooth prediction model includes a semantic segmentation model and an object detection model for parallel processing of images, and also includes a screening module for combining the tooth region results output by the semantic segmentation model and the object detection model and screening out error regions;
[0009] Train the oral tooth prediction model using the training sample;
[0010] Input the oral medical image to be measured into the trained oral tooth prediction model for recognition and segmentation, output the tooth region of the oral medical image to be measured and encode it for representation.
[0011] Preferably, in the above tooth region detection method provided by the embodiments of the present invention, the preprocessing of the oral medical image dataset includes:
[0012] Perform data augmentation processing on the oral medical image dataset.
[0013] Preferably, in the above tooth region detection method provided by the embodiments of the present invention, the preprocessing of the oral medical image dataset further includes:
[0014] Use the interpolation method to adjust the image size in the oral medical image dataset to a unified size;
[0015] Use the high-contrast retention algorithm to process the resized image to obtain high-contrast retention features;
[0016] Normalize the processed image.
[0017] Preferably, in the above tooth region detection method provided by the embodiments of the present invention, the screening module is specifically used to combine the tooth region output by the semantic segmentation model, the tooth region detection box and tooth region coordinates output by the object detection model, screen out the wrong regions from the combination result to obtain the pseudo-label of the training sample, and add unlabeled oral medical image data; use the other tooth regions in the combination result except the wrong regions as the screened tooth regions.
[0018] Preferably, in the above tooth region detection method provided by the embodiments of the present invention, the oral tooth prediction model further includes a correction module for correcting the tooth regions screened by the screening module and updating the labeled oral medical image data.
[0019] Preferably, in the above tooth region detection method provided by the embodiments of the present invention, the semantic segmentation model is a deep learning neural network model including four convolutional layers, four pooling layers, four activation layers, four pooling layers, four upsampling layers and one output layer; the semantic segmentation model uses a fully connected conditional random field;
[0020] The object detection model is a deep learning neural network model including thirteen convolutional layers, fifteen activation layers, four pooling layers, two fully connected layers and one output layer.
[0021] Preferably, in the above tooth region detection method provided by the embodiments of the present invention, it further includes:
[0022] Use the trained oral tooth prediction model to predict and analyze unlabeled oral medical images, analyze independent tooth regions from the prediction results according to preset rules, and mark the independent tooth regions.
[0023] An embodiment of the present invention further provides a tooth region detection device, including:
[0024] A dataset acquisition module for acquiring an oral medical image dataset; the oral medical image dataset includes oral medical images and oral medical image data labeled on the oral medical images;
[0025] A dataset preprocessing module for preprocessing the oral medical image dataset and using the preprocessed oral medical image dataset as training samples;
[0026] A model construction module for constructing an oral tooth prediction model; the oral tooth prediction model includes a semantic segmentation model and an object detection model for parallel image processing, and further includes a screening module for combining the tooth region results output by the semantic segmentation model and the object detection model and screening out error regions;
[0027] A model training module for training the oral tooth prediction model using the training samples;
[0028] A tooth region recognition module for inputting a to-be-detected oral medical image into the trained oral tooth prediction model for recognition and segmentation, outputting the tooth region of the to-be-detected oral medical image, and encoding and representing it.
[0029] An embodiment of the present invention further provides a tooth region detection device, including a processor and a memory. Among them, when the processor executes a computer program stored in the memory, the above-mentioned tooth region detection method provided by the embodiment of the present invention is implemented.
[0030] An embodiment of the present invention further provides a computer-readable storage medium for storing a computer program. Among them, when the computer program is executed by a processor, the above-mentioned tooth region detection method provided by the embodiment of the present invention is implemented.
[0031] As can be seen from the above technical solution, a method for detecting tooth regions provided by the present invention includes: obtaining an oral medical image dataset; the oral medical image dataset includes oral medical images and oral medical image data annotated on the oral medical images; preprocessing the oral medical image dataset and using the preprocessed oral medical image dataset as training samples; constructing an oral tooth prediction model; the oral tooth prediction model includes a semantic segmentation model and an object detection model for parallel processing of images, and further includes a screening module for combining the tooth region results output by the semantic segmentation model and the object detection model and screening out incorrect regions; training the oral tooth prediction model using the training samples; inputting the oral medical image to be measured into the trained oral tooth prediction model for recognition and segmentation, outputting the tooth region of the oral medical image to be measured and representing it in coding.
[0032] The present invention collects oral medical images to train the semantic segmentation model and the object detection model, integrates the semantic segmentation model and the object detection model into one framework, and processes oral medical images using the idea of parallelism. For the input oral medical images, the semantic segmentation and object detection models will simultaneously read and process the images. The screening module combines the output results of the two models and, after removing incorrect regions, can effectively obtain accurately recognized and clearly segmented tooth regions. The introduction of the parallel processing framework based on the semantic segmentation model and the object detection model makes the image segmentation process more accurate, compensates for the defect of inaccurate segmentation and positioning of a single model, reduces the human interference factors in the film reading process and the dependence on the professional quality of doctors, and further improves the accuracy, consistency and credibility.
[0033] In addition, the present invention also provides corresponding devices, equipment and computer-readable storage media for the tooth region detection method, further making the above method more practical, and the devices, equipment and computer-readable storage media have corresponding advantages. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are only embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.
[0035] Figure 1 It is a flowchart of the tooth region detection method provided by the embodiment of the present invention;
[0036] Figure 2 It is a schematic structural diagram of the tooth region detection device provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0037] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all 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.
[0038] The present invention provides a method for detecting a tooth region, as Figure 1 shown, which includes the following steps:
[0039] S101. Obtain an oral medical image dataset; the oral medical image dataset includes oral medical images and oral medical image data annotated on the oral medical images;
[0040] This step is the original image acquisition stage. The oral medical image data here can be annotated by professional doctors, and the oral medical images can be panoramic oral tooth images, which contain complete oral tooth structure information. Preferably, the selection of oral medical images is random, including experimental subjects of different genders and age groups.
[0041] S102. Preprocess the oral medical image dataset and use the preprocessed oral medical image dataset as training samples;
[0042] This step is the preprocessing stage. There are many preprocessing methods, which can be selected according to the actual situation. Specifically, each oral medical image and the annotated oral medical image form a training sample, and multiple training samples form a dataset; the dataset is divided into a training set, a validation set, and a test set; the training set is used for model training, the validation set is used for model adjustment, and the test set is for model testing; the division ratio of the training set to the test set is 4:1. For example, there are 1000 samples in the training set and 200 samples in the test set.
[0043] S103. Build an oral tooth prediction model; the oral tooth prediction model includes a semantic segmentation model and an object detection model for parallel processing of images, and also includes a screening module for combining the tooth region results output by the semantic segmentation model and the object detection model and screening out the error regions;
[0044] This step is the model construction stage, and an image parallel processing framework is mainly used as the classifier structure. The work of semantic segmentation is to classify each pixel point in the image and determine the category of each pixel point, for example, belonging to teeth, gums or other parts, etc., so as to achieve the division of different regions. The image semantic segmentation model for the segmentation of oral medical images, compared with the instance segmentation model, only distinguishes the types of pixel points, but does not distinguish the tooth instances to which the pixel points belong. The work of object detection is to classify the objects in the image and identify the positions of the objects, for example, the positions of tooth objects, gum objects or other object positions, etc., so as to achieve the recognition of different regions.
[0045] S104. Train the oral tooth prediction model using the training samples;
[0046] This step is the training stage. The training method of the semantic segmentation model includes: inputting the training samples into the semantic segmentation model to obtain the tooth region. The training method of the object detection model includes: inputting the training samples into the object detection model to obtain the tooth region detection frame and the tooth region coordinates.
[0047] S105. Input the oral medical image to be measured into the trained oral tooth prediction model for recognition and segmentation, output the tooth region of the oral medical image to be measured and represent it by coding;
[0048] This step is the prediction stage. The trained oral tooth prediction model is used to recognize and segment the oral medical image, and all tooth regions are represented by coding.
[0049] In the above tooth region detection method provided by the embodiments of the present invention, oral medical images are collected to train the semantic segmentation model and the object detection model, the semantic segmentation model and the object detection model are integrated into a framework, and the idea of parallelism is adopted to process oral medical images. For the input oral medical images, the semantic segmentation and object detection models will read and process the images simultaneously. The screening module combines the output results of the two models, and after removing the error regions, the tooth regions with accurate recognition and clear segmentation can be effectively obtained. The introduction of the parallel processing framework based on the semantic segmentation model and the object detection model makes the image segmentation process more accurate, makes up for the defect of inaccurate segmentation and positioning of a single model, reduces the artificial interference factors in the film reading process and the dependence on the professional quality of doctors, and further improves the accuracy, consistency and credibility.
[0050] In specific implementation, in the above tooth region detection method provided by the embodiments of the present invention, step S101 for preprocessing the oral medical image dataset may include: performing data augmentation processing on the oral medical image dataset, such as horizontal flipping and vertical flipping, etc., which can enrich the oral medical image training set, reduce the phenomenon of model overfitting, and enhance the generalization ability of the model.
[0051] Further, the preprocessing of the oral medical image dataset in step S101 may further include: using an interpolation method to adjust the image size in the oral medical image dataset to a unified size, which can better extract image features and prevent model overfitting. Specifically, oral medical images usually have a large size. The labeled oral medical images are scaled, and the images are scaled to a preset 512*512. The phenomenon of model overfitting is reduced, and the generalization ability of the model is enhanced.
[0052] In practical applications, random cropping can be performed on the image and the annotation map. For the case where the cropping size is larger than the original image, a filling operation is performed to fill the original image to the required cropping size.
[0053] Further, the preprocessing of the oral medical image dataset in step S101 may further include: using a high-contrast retention algorithm to process the resized image to obtain high-contrast retention features. Specifically, a Gaussian filter is used to smooth the image, and then the original image is subtracted from the Gaussian-blurred image to obtain an enhanced edge value. On the basis of the blurred image, the enhanced edge value multiplied by a correlation coefficient is added to obtain an image with enhanced effect, reducing image noise and retaining edge detail features. Specifically, to avoid most pixel values exceeding the effective range and losing image information, 127 is added to the enhanced edge value obtained by subtracting the Gaussian-blurred image from the original image to obtain high-contrast retention features. The Gaussian distribution equation in N-dimensional space is:
[0054]
[0055] In the formula, r represents the blur radius, σ represents the standard deviation of the normal distribution, and N represents the dimension of the sample.
[0056] The Gaussian distribution equation in two-dimensional space is:
[0057]
[0058] In the formula, u represents the x coordinate of the sample, and v represents the y coordinate of the sample.
[0059] Further, the preprocessing of the oral medical image dataset in step S101 may further include: normalizing the processed image. This can maintain affine invariance, simplify the computational complexity of the model, and accelerate the convergence of training the neural network. Specifically, the range transformation method is used to convert the pixel values of the oral medical image into decimals between 0 and 1, eliminating the influence of dimension and data value range. The dimensional expression is converted into a dimensionless expression to become a scalar. The normalization formula is as follows:
[0060]
[0061] In the formula, \(x\) represents the entire sample set, and \(x\) i represents the \(i\)-th sample, and norm represents the sample set after normalization.
[0062] In specific implementation, in the above-mentioned tooth region detection method provided by the embodiments of the present invention, the screening module is specifically used to combine the tooth region output by the semantic segmentation model, the tooth region detection frame and the tooth region coordinates output by the target detection model, screen out the error regions from the combination result to obtain the pseudo-label of the training sample, and add the unlabeled oral medical image data; use the other tooth regions except the error regions in the combination result as the screened tooth regions.
[0063] In specific implementation, in the above-mentioned tooth region detection method provided by the embodiments of the present invention, the oral tooth prediction model further includes: a correction module for correcting the tooth regions screened by the screening module and updating the labeled oral medical image data. Correct the pixel points and coordinates of the identified regions to achieve a pixel-level discrimination effect
[0064] In specific implementation, in the above-mentioned tooth region detection method provided by the embodiments of the present invention, the semantic segmentation model may include a convolutional layer, an activation layer, a pooling layer, an upsampling layer, and a softmax layer. Preferably, the semantic segmentation model can be set as a deep learning neural network model including four convolutional layers, four pooling layers, four activation layers, four pooling layers, four upsampling layers, and one output layer. The semantic segmentation model uses a fully connected conditional random field. The conditional random field used by the semantic segmentation model conforms to the Gibbs distribution:
[0065]
[0066] In the formula, \(E(x|I)\) represents the energy function, \(Z(I)\) represents the normalization factor, and \(P(X = x|I)\) represents the conditional probability.
[0067]
[0068] In the formula, the unary potential function comes from the output of the previous fully connected network, and \(i, j\) respectively represent represents the unary potential energy, which is only related to the value of point \(i\), represents the binary potential energy, the relationship between each node, and \(E(x)\) represents the potential energy function.
[0069] In specific implementation, in the above-mentioned tooth region detection method provided by the embodiments of the present invention, the target detection model may include a convolutional layer, an activation layer, a pooling layer, a fully connected layer, and a softmax layer. Preferably, the target detection model may be set as a deep learning neural network model including thirteen convolutional layers, fifteen activation layers, four pooling layers, two fully connected layers, and an output layer. Among them, the neural network parameter update formula is:
[0070]
[0071]
[0072] In the formula, W represents the weight, b represents the bias, z represents the output result of each layer of the neural network, and l represents the l-th layer of the neural network; represents the gradient of the weight W of the l-th layer, the partial derivative of the loss function with respect to W, and the gradient represents the fastest change of the function in the gradient direction at a point; represents the gradient of the bias b of the l-th layer, the partial derivative of the loss function with respect to b; represents the partial derivative of the loss function with respect to the output function z of the l-th output layer; represents the partial derivative of the output function of the l-th layer with respect to the weight W of the l-th layer; represents the partial derivative of the output function of the l-th layer with respect to the bias b of the l-th layer.
[0073] Furthermore, in specific implementation, in the above-mentioned tooth region detection method provided by the embodiments of the present invention, it may further include: using the trained oral tooth prediction model to predict and analyze the unlabeled oral medical images, analyzing the independent tooth regions from the prediction results according to the preset rules, and marking the independent tooth regions.
[0074] Specifically, the preset marking rules are set according to the dental standard tooth position notation method. Among them, the tooth position notation method is a method of numbering each human tooth; the upper and lower tooth rows are divided into four regions of upper, lower, left, and right by a cross symbol. The upper right region is also called region A, the upper left region is also called region B, the lower right region is also called region C, and the lower left region is also called region D. Each tooth is marked with two Arabic numerals. The first digit represents the quadrant where the tooth is located: the upper right, upper left, lower left, and lower right of the patient are 1, 2, 3, and 4 in permanent teeth and 5, 6, 7, and 8 in deciduous teeth; the second digit represents the position of the tooth: from the central incisor to the third molar is 1-8.
[0075] The following uses an example to illustrate the specific steps of the above-mentioned tooth region detection method provided by the embodiments of the present invention:
[0076] Step 1. Data acquisition. 2834 samples are collected in the oral hospital, and each sample includes the original oral medical image data and the corresponding labeled oral medical image data of the original oral medical image data.
[0077] Step 2: Preprocessing of the image. Operations such as flipping, scaling, Gaussian blurring, and normalization are performed on the image to enrich the training set of oral medical images, reduce the phenomenon of model overfitting, and enhance the generalization ability of the model.
[0078] Step 3: Division of the sample set. According to the concentration gradient method, 2834 experimental samples are divided into a training set consisting of 2267 samples and a test set consisting of 567 samples at a ratio of 4:1.
[0079] Step 4: Establishment of the model. The parallel processing framework mainly includes a semantic segmentation model and an object detection model. The preset semantic segmentation model is a deep learning neural network model including four convolutional layers, four pooling layers, four activation layers, four pooling layers, four upsampling layers, and one output layer. The preset object detection model is a deep learning neural network model including thirteen convolutional layers, fifteen activation layers, four pooling layers, two fully connected layers, and one output layer.
[0080] Step 5: Model testing. Predictions are made on the modeling set and the prediction set, and the samples with a threshold exceeding 0.9 in the output results are recorded as positive samples. The selected model has an accuracy of 0.96 on the test set, obtaining a satisfactory prediction accuracy.
[0081] By performing the above-mentioned steps 1 to 5, the tooth area in the oral image can be effectively identified, reducing the artificial interference factors and the dependence on the professional quality of doctors during the film reading process.
[0082] Based on the same inventive concept, an embodiment of the present invention further provides a tooth area detection device. Since the principle of solving problems by this device is similar to that of the aforementioned tooth area detection method, the implementation of this device can refer to the implementation of the tooth area detection method, and the repeated parts will not be described again.
[0083] In specific implementation, the tooth area detection device provided by the embodiment of the present invention, as Figure 2 shown, specifically includes:
[0084] A dataset acquisition module 11 for acquiring an oral medical image dataset; the oral medical image dataset includes oral medical images and oral medical image data annotated on the oral medical images;
[0085] A dataset preprocessing module 12 for preprocessing the oral medical image dataset and using the preprocessed oral medical image dataset as training samples;
[0086] A model construction module 13 for constructing an oral tooth prediction model; the oral tooth prediction model includes a semantic segmentation model and an object detection model for parallel processing of images, and also includes a screening module for combining the tooth region results output by the semantic segmentation model and the object detection model and screening out incorrect regions;
[0087] A model training module 14 for training the oral tooth prediction model using training samples;
[0088] A tooth region recognition module 15 for inputting a to-be-detected oral medical image into the trained oral tooth prediction model for recognition and segmentation, outputting the tooth region of the to-be-detected oral medical image and encoding it.
[0089] In the above-mentioned tooth region detection device provided by the embodiment of the present invention, through the interaction of the above five modules, for the input oral medical image, the semantic segmentation and object detection models will simultaneously read and process the image, and the screening module combines the output results of the two models. After removing the incorrect regions, a tooth region with accurate recognition and clear segmentation can be effectively obtained, making up for the defect of inaccurate segmentation and positioning of a single model, reducing the human interference factors and the dependence on the professional quality of doctors during the film reading process, and further improving the accuracy, consistency and credibility.
[0090] In specific implementation, in the above-mentioned tooth region detection method provided by the embodiment of the present invention, it may further include: an independent tooth region marking module for using the trained oral tooth prediction model to predict and analyze unlabeled oral medical images, analyzing independent tooth regions from the prediction results according to preset rules, and marking the independent tooth regions.
[0091] For the more specific working processes of the above-mentioned various modules, reference may be made to the corresponding content disclosed in the foregoing embodiments, and details will not be elaborated herein.
[0092] Correspondingly, the embodiment of the present invention also discloses a tooth region detection device, including a processor and a memory; wherein, when the processor executes the computer program stored in the memory, the tooth region detection method disclosed in the foregoing embodiments is implemented.
[0093] For the more specific process of the above method, reference may be made to the corresponding content disclosed in the foregoing embodiments, and details will not be elaborated herein.
[0094] Further, the present invention also discloses a computer-readable storage medium for storing a computer program; when the computer program is executed by a processor, the tooth region detection method disclosed above is implemented.
[0095] For the more specific process of the above method, reference may be made to the corresponding content disclosed in the foregoing embodiments, and details will not be elaborated herein.
[0096] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the embodiments, reference can be made to each other. For the devices, equipment, and storage media disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For the relevant parts, reference can be made to the description in the method section.
[0097] Those skilled in the art can further realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed in this article can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of the examples have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.
[0098] The steps of the methods or algorithms described in combination with the embodiments disclosed in this article can be directly implemented by hardware, software modules executed by a processor, or a combination of the two. The software modules can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium well-known in the technical field.
[0099] In summary, a tooth region detection method provided by an embodiment of the present invention includes: obtaining an oral medical image data set; the oral medical image data set includes oral medical images and oral medical image data annotated on the oral medical images; preprocessing the oral medical image data set, and using the preprocessed oral medical image data set as a training sample; constructing an oral tooth prediction model; the oral tooth prediction model includes a semantic segmentation model and an object detection model for parallel processing of images, and further includes a screening module for combining the tooth region results output by the semantic segmentation model and the object detection model and screening out incorrect regions; training the oral tooth prediction model using the training sample; inputting the oral medical image to be measured into the trained oral tooth prediction model for recognition and segmentation, outputting the tooth region of the oral medical image to be measured and performing coding representation. In this way, oral medical images are collected to train the semantic segmentation model and the object detection model, the semantic segmentation model and the object detection model are integrated into one framework, and the idea of parallel processing is adopted to process oral medical images. For the input oral medical image, the semantic segmentation and object detection models will simultaneously read and process the image. The screening module combines the output results of the two models, and after removing the incorrect regions, an accurately recognized and clearly segmented tooth region can be effectively obtained. The introduction of the parallel processing framework based on the semantic segmentation model and the object detection model makes the image segmentation process more accurate, compensates for the defect of inaccurate segmentation and positioning of a single model, reduces the artificial interference factors and the dependence on the professional quality of doctors in the film reading process, and further improves the accuracy, consistency and credibility. In addition, the present invention also provides a corresponding device, equipment and computer-readable storage medium for the tooth region detection method, further making the above method more practical, and the device, equipment and computer-readable storage medium have corresponding advantages.
[0100] Finally, it should also be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.
[0101] The above has introduced in detail the tooth area detection method, device, equipment and storage medium provided by the present invention. Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.
Claims
1. A method for detecting a tooth area, characterized in that, Including: Obtain an oral medicine image dataset; the oral medicine image dataset includes oral medicine images and oral medicine image data annotated on the oral medicine images; Preprocess the oral medicine image dataset, and use the preprocessed oral medicine image dataset as a training sample; Construct an oral tooth prediction model; the oral tooth prediction model includes a semantic segmentation model and an object detection model for parallel processing of images, and also includes a screening module for combining the tooth region results output by the semantic segmentation model and the object detection model and screening out incorrect regions; specifically, the screening module is used to combine the tooth regions output by the semantic segmentation model, the tooth region detection frames and tooth region coordinates output by the object detection model, screen out incorrect regions from the combined results to obtain pseudo-labels of the training sample, and add unannotated oral medicine image data; Use other tooth regions in the combined results except for the incorrect regions as the screened tooth regions; Train the oral tooth prediction model using the training sample; Input the to-be-detected oral medicine image into the trained oral tooth prediction model for recognition and segmentation, output the tooth region of the to-be-detected oral medicine image and perform encoding representation.
2. The tooth region detection method according to claim 1, characterized in that The preprocessing of the oral medicine image dataset includes: Perform data augmentation processing on the oral medicine image dataset.
3. The tooth region detection method according to claim 2, wherein The preprocessing of the oral medicine image dataset further includes: Use an interpolation method to adjust the image size in the oral medicine image dataset to a unified size; Use a high-contrast retention algorithm to process the image after size adjustment to obtain high-contrast retention features; Normalize the processed image.
4. The tooth area detection method according to claim 1, wherein The oral tooth prediction model further includes: a correction module for correcting the tooth regions screened by the screening module and updating the annotated oral medicine image data.
5. The tooth area detection method according to claim 1, wherein The semantic segmentation model is a deep learning neural network model including four convolutional layers, four pooling layers, four activation layers, four pooling layers, four upsampling layers and an output layer; the semantic segmentation model uses a fully connected conditional random field; The object detection model is a deep learning neural network model including thirteen convolutional layers, fifteen activation layers, four pooling layers, two fully connected layers and an output layer.
6. The tooth region detection method according to claim 1, characterized in that, Also including: Use the trained oral tooth prediction model to predict and analyze unannotated oral medicine images, analyze independent tooth regions from the prediction results according to preset rules, and mark the independent tooth regions.
7. A dental area detection device, characterized in that, Including: A dataset acquisition module for obtaining an oral medicine image dataset; the oral medicine image dataset includes oral medicine images and oral medicine image data annotated on the oral medicine images; A dataset preprocessing module for preprocessing the oral medicine image dataset and using the preprocessed oral medicine image dataset as a training sample; A model construction module for constructing an oral tooth prediction model; the oral tooth prediction model includes a semantic segmentation model and an object detection model for parallel processing of images, and further includes a screening module for combining the tooth region results output by the semantic segmentation model and the object detection model and screening out incorrect regions; the screening module is specifically configured to combine the tooth regions output by the semantic segmentation model, the tooth region detection frames and tooth region coordinates output by the object detection model, screen out incorrect regions from the combined result to obtain pseudo-labels of the training samples, and add unlabeled oral medical image data; Regarding other tooth regions in the combined result except for the incorrect regions as the screened tooth regions; A model training module for training the oral tooth prediction model using the training samples; A tooth region recognition module for inputting a to-be-tested oral medical image into the trained oral tooth prediction model for recognition and segmentation, outputting the tooth region of the to-be-tested oral medical image and performing encoding representation.
8. A dental area detection device, characterized in that, It includes a processor and a memory. Among them, when the processor executes the computer program stored in the memory, the tooth region detection method according to any one of claims 1 to 6 is implemented.
9. A computer-readable storage medium, characterized in that, For storing a computer program, wherein when the computer program is executed by a processor, the tooth region detection method according to any one of claims 1 to 6 is implemented.