Alveolar Bone Resorption Judgment Method Based on SegFormer and Orthopantomogram

Abstract

The present invention discloses a method for judging alveolar bone resorption based on SegFormer and panoramic dental films, which includes a training step of a tooth recognition and segmentation model and a step of judging alveolar bone resorption. Among them, the training step includes: obtaining multiple panoramic dental films and marking the tooth regions; segmenting the tooth regions; performing data augmentation on the alveolar bone resorption data; segmenting the crowns of teeth, supra-bony roots and sub-bony roots of teeth in the alveolar bone resorption data after data augmentation. The absorption judgment step includes: using the tooth recognition and segmentation model to segment the tooth regions from the panoramic dental film to be judged for alveolar bone resorption, then segmenting the crowns of teeth, supra-bony roots and sub-bony roots of teeth in the tooth regions, and then calculating to obtain the alveolar bone resorption judgment result. The present invention performs visualization processing of the segmentation results and generates corresponding alveolar bone resorption measurement values, simplifies the interpretation process of panoramic dental films, and provides the possibility for further operations to generate more accurate diagnoses.

Description

Technical Field

[0001] The present invention relates to oral disease diagnosis technology, specifically a method for judging alveolar bone resorption based on SegFormer and orthopantomogram. Background Art

[0002] Periodontitis is one of the most common oral diseases and the main cause of tooth loss in adults. Judging the degree of alveolar bone resorption through imaging examination is an important auxiliary examination method for the diagnosis of periodontal diseases. As a commonly used imaging examination method in the clinical diagnosis and treatment process of periodontics, the orthopantomogram has the advantages of relatively low price and certain repeatability. When interpreting the results of orthopantomogram examination, the cementoenamel junction is used as the demarcation point between the root and the crown of the tooth, and the height of the alveolar bone at the mesial and distal margins of the tooth is used as the alveolar bone level point. Based on the diagnostic criteria of the 2018 International New Classification of Periodontal and Peri-implant Diseases, the percentage of radiographic alveolar bone resorption occupying the root length and its ratio to age are used as one of the diagnostic criteria for staging and grading periodontitis, where the bone resorption percentage < 15% is stage I, 15 - 33% is stage II, > 33% is stage III or IV; the ratio of the bone resorption percentage to age < 0.25 is grade A, 0.25 - 1 is grade B, > 1 is grade C. Physiologically, there is a distance of 1 - 2 mm between the alveolar bone level and the cementoenamel junction, and within this range, it can be considered that there is no obvious alveolar bone resorption.

[0003] The traditional method for identifying and judging alveolar bone resorption from orthopantomograms is manual judgment. Due to the subjectivity of the manual judgment results, their quality is restricted by the learning level and experience of the film readers, and there are differences among different film readers. The time required for accurate manual measurement of the degree of alveolar bone resorption is relatively long, and its feasibility is low in clinical practice. Therefore, estimation is usually used in actual clinical work. However, estimation is not conducive to accurate diagnosis at the staging critical value, and it is also impossible to quickly judge the tooth position with the most severe resorption in the case of relatively average alveolar bone resorption in the whole mouth to obtain accurate diagnostic indicators. Summary of the Invention

[0004] The purpose of the present invention is to overcome the problems of strong subjectivity, long time consumption, and inaccurate interpretation results in the diagnosis of periodontal diseases by the existing manual film reading method, and provide a method for judging alveolar bone resorption based on SegFormer and orthopantomogram. When applied, it can identify the crown, supra-bony part of the root, and sub-bony part of the root of the tooth in the orthopantomogram, perform visual processing of the segmentation results, and generate corresponding alveolar bone resorption measurement values, simplifying the process of interpreting orthopantomograms, reducing the technical sensitivity of manual film reading, and providing the possibility for further operations to generate more accurate diagnoses.

[0005] The purpose of the present invention is mainly achieved through the following technical solutions:

[0006] An alveolar bone resorption judgment method based on SegFormer and panoramic dental films, including a tooth recognition and segmentation model training step for training a tooth recognition and segmentation model, and an alveolar bone resorption judgment step for judging the alveolar bone resorption of each tooth in a panoramic dental film based on the tooth recognition and segmentation model. The tooth recognition and segmentation model includes a first SegFormer model and a second SegFormer model. The first SegFormer model is used to segment the tooth area in the panoramic dental film, and the second SegFormer model is used to segment the crown, supra-bony root and sub-bony root of the tooth in the panoramic dental film that only contains the tooth area. The tooth recognition and segmentation model training step includes the following steps:

[0007] Step S11: Obtain multiple panoramic dental films and mark the tooth areas of each panoramic dental film by color filling, and use all the marked panoramic dental films as the first alveolar bone resorption data to form a first alveolar bone resorption dataset. Among them, the tooth area includes the crown, supra-bony root and sub-bony root, and different colors are used to fill the crown, supra-bony root and sub-bony root.

[0008] Step S12: Use the first SegFormer model to segment the tooth areas of all panoramic dental films, and remove the non-tooth areas in the panoramic dental films to obtain panoramic dental films that only contain the tooth areas as the second alveolar bone resorption data to form a second alveolar bone resorption dataset.

[0009] Step S13: Perform data augmentation on the second alveolar bone resorption data in the second alveolar bone resorption dataset.

[0010] Step S14: Use the second SegFormer model to segment the crown, supra-bony root and sub-bony root of the teeth in the second alveolar bone resorption data after data augmentation to obtain the segmentation result.

[0011] The alveolar bone resorption judgment step includes the following steps:

[0012] Step S21: Use the first SegFormer model to segment the tooth area of the panoramic dental film to be judged for alveolar bone resorption, and remove the non-tooth area.

[0013] Step S22: Use the second SegFormer model to segment the crown, supra-bony root and sub-bony root of the teeth in the tooth area to obtain the segmentation result.

[0014] Step S23: Calculate the percentage of alveolar bone resorption of each tooth in the panoramic dental film accounting for the root length according to the segmentation result, and obtain the alveolar bone resorption judgment result of each tooth. When the present invention is applied, for the training step of the tooth recognition and segmentation model to improve the training accuracy, multiple panoramic dental films are adopted, and the alveolar bone resorption judgment step processes a single panoramic dental film.

[0015] Further, when marking in step S11, the cementoenamel junction points at the mesial and distal of the tooth are used as the demarcation points between the crown and the root of the tooth, and the alveolar bone level points at the mesial and distal of the tooth are used as the demarcation points between the supra-bony part of the root and the sub-bony part of the root.

[0016] Further, when marking in step S11, the wisdom teeth and the teeth that have lost the cementoenamel junction points on the panoramic dental film are not marked; when it is impossible to judge the cementoenamel junction point due to tooth repair or filling treatment, the root margin of the restoration or filling on the tooth contour is used as the cementoenamel junction point.

[0017] Further, the first SegFormer model includes two Transformer encoding modules. The steps of using the first SegFormer model to segment the tooth regions of all panoramic dental films in step S12 include the following steps:

[0018] Take the circumscribed rectangle of the first alveolar bone resorption dataset label as the tooth region label to obtain the tooth region dataset, and use the first SegFormer model including two Transformer encoding modules to train the tooth region dataset to segment the tooth regions in the panoramic dental film.

[0019] Further, the data augmentation methods in step S13 include horizontal flipping, vertical flipping and contrast transformation.

[0020] Further, the second SegFormer model includes four Transformer encoding modules.

[0021] Further, the percentage of alveolar bone resorption accounting for the root length in step S23 is obtained by calculating the ratio of the distance between the cementoenamel junction and the alveolar bone crest to the distance between the cementoenamel junction and the apex point according to the segmentation result.

[0022] In summary, the present invention has the following beneficial effects compared with the prior art:

[0023] (1) When the present invention is applied, in the model training stage, first, the tooth crown, the supra-bony part of the tooth root, and the infra-bony part of the tooth root in the panoramic dental film are marked to make a data set; then, the first SegFormer model is used to segment the tooth region in the panoramic dental film, removing useless information to train the first SegFormer model; then, data augmentation is performed on the alveolar bone resorption data set containing only the tooth region; finally, the second SegFormer model is used to segment the tooth crown, the supra-bony part of the tooth root, and the infra-bony part of the tooth root of the teeth in the data-augmented alveolar bone resorption data to train the second SegFormer model. When the present invention judges the alveolar bone resorption of each tooth in the panoramic dental film, the trained first SegFormer model is used to segment the tooth region of the panoramic dental film to be judged for alveolar bone resorption, and then the trained second SegFormer model is used to segment the tooth crown, the supra-bony part of the tooth root, and the infra-bony part of the tooth root of the teeth in the tooth region. Then, relevant numerical calculations are performed according to the segmentation results by tooth position to obtain the percentage of bone resorption in the root length, and a preliminary diagnosis of periodontitis is further carried out.

[0024] (2) When the present invention is applied, the SegFormer model can achieve high-precision segmentation of alveolar bone resorption in panoramic dental films, optimizing the traditional panoramic dental film diagnosis process and output results, saving a large amount of labor costs, and reducing the technical sensitivity of periodontal disease diagnosis based on panoramic dental films. Description of the Drawings

[0025] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, form a part of this application, and do not limit the embodiments of the present invention. In the drawings:

[0026] Figure 1 is a flowchart of the training steps of the tooth recognition and segmentation model in a specific embodiment of the present invention;

[0027] Figure 2 is a flowchart of the alveolar bone resorption judgment steps in a specific embodiment of the present invention;

[0028] Figure 3 is a schematic diagram of the distribution of the cementoenamel junction at the mesial and distal sides of tooth 25, and the alveolar crest levels at the mesial and distal sides;

[0029] Figure 4 is a schematic diagram of the annotation of the alveolar bone resorption data set of the panoramic dental film when the present invention is applied in a specific embodiment;

[0030] Figure 5 is a flowchart of the image processing of the alveolar bone resorption data set corresponding to the tooth recognition and segmentation model training stage when the present invention is applied in a specific embodiment. Detailed Embodiments

[0031] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in combination with embodiments and drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and do not limit the present invention.

[0032] Embodiment:

[0033] As Figure 1 and Figure 2 shown, the alveolar bone resorption judgment method based on SegFormer and panoramic dental radiographs includes a training step of a tooth recognition and segmentation model for training the tooth recognition and segmentation model, and an alveolar bone resorption judgment step of judging the alveolar bone resorption of each tooth in the panoramic dental radiograph based on the tooth recognition and segmentation model. Among them, the tooth recognition and segmentation model of this embodiment includes a first SegFormer model and a second SegFormer model. The first SegFormer model is used to segment the tooth area in the panoramic dental radiograph, and the second SegFormer model is used to segment the crown, supra-bony root, and sub-bony root of the tooth in the panoramic dental radiograph that only contains the tooth area.

[0034] The training step of the tooth recognition and segmentation model of this embodiment includes the following steps: Step S11, obtain multiple panoramic dental radiographs and mark the tooth areas of each panoramic dental radiograph by means of color filling, and use all the marked panoramic dental radiographs as the first alveolar bone resorption data to form the first alveolar bone resorption data set; among them, the tooth area includes the crown, supra-bony root, and sub-bony root, and different colors are used to fill the crown, supra-bony root, and sub-bony root; Step S12, use the first SegFormer model to segment the tooth areas of all panoramic dental radiographs, remove the non-tooth areas in the panoramic dental radiograph, and obtain the panoramic dental radiograph that only contains the tooth area as the second alveolar bone resorption data to form the second alveolar bone resorption data set; Step S13, perform data augmentation on the second alveolar bone resorption data in the second alveolar bone resorption data set; Step S14, use the second SegFormer model to segment the crown, supra-bony root, and sub-bony root of the tooth in the second alveolar bone resorption data after data augmentation to obtain the segmentation result.

[0035] The steps for judging alveolar bone resorption in this embodiment include the following steps: Step S21, segment the dental panoramic radiograph of the alveolar bone to be judged by using the first SegFormer model to remove the non-dental area; Step S22, segment the crown, supra-bony root part and sub-bony root part of the teeth in the dental area by using the second SegFormer model to obtain the segmentation result; Step S23, calculate the percentage of alveolar bone resorption of each tooth in the dental panoramic radiograph accounting for the root length to obtain the alveolar bone resorption judgment result of each tooth. Among them, the percentage of alveolar bone resorption accounting for the root length is obtained by calculating the ratio of the distance between the cementoenamel junction and the alveolar bone crest top to the distance between the cementoenamel junction and the apex point according to the segmentation result. The segmentation results obtained in Step S14 and Step S22 in this embodiment are the crown, supra-bony root part and sub-bony root part of the teeth.

[0036] In this embodiment, the first alveolar bone resorption data in the first alveolar bone resorption data set is marked by the method of manual annotation using existing mapping software such as PHOTOSHOP. When marking, the cementoenamel junction points on the mesial and distal sides of the tooth are used as the demarcation points between the crown and the root, and the alveolar bone level points on the mesial and distal sides of the tooth are used as the demarcation points between the supra-bony root part and the sub-bony root part, that is, the crown and the root are segmented by the cementoenamel junction points on the mesial and distal sides of the tooth, and the supra-bony part and the sub-bony part of the root are segmented by the alveolar bone level points on the mesial and distal sides of the tooth. Among them, the cementoenamel junction refers to the position where the enamel of the tooth is adjacent to the cementum. The alveolar bone level point refers to the highest point of the alveolar bone crest on the mesial or distal side of the tooth. As Figure 3 shown, M1 and D1 respectively represent the cementoenamel junctions on the mesial and distal sides of tooth 25, and M2 and D2 respectively represent the alveolar level points on the mesial and distal sides of tooth 25.

[0037] In this embodiment, the entire tooth is marked according to the contour of the tooth, and these three parts are filled with different colors respectively. Judging alveolar bone resorption is for diagnosing periodontitis, and wisdom teeth are not within the scope of this diagnosis. In addition, many wisdom teeth are buried in the alveolar bone and it is impossible to judge alveolar bone resorption. Therefore, all wisdom teeth are not marked when this embodiment is applied. When the tooth is a residual root and has lost the cementoenamel junction, firstly, it is impossible to confirm the position of the cementoenamel junction, and secondly, losing the cementoenamel junction usually means that the residual root has no retention value and marking has no meaning. Therefore, in this embodiment, when the tooth is a residual root and has lost the cementoenamel junction point, it is not marked. When the tooth has undergone restoration or filling treatment and it is impossible to judge the cementoenamel junction point, the root-side edge of the restoration or filling on the tooth contour is used as the cementoenamel junction point. A two-dimensional plane tooth has four directions: the crown side, the root side, the mesial side and the distal side. The root-side edge in using the root-side edge of the restoration or filling on the tooth contour as the cementoenamel junction point refers to the edge of the restoration or filling close to the root direction.

[0038] Figure 4 An example of manual marking for alveolar bone resorption data in panoramic dental radiographs Figure 4 (a) is the unmarked image Figure 4 (b) is the image after marking. The purpose of manual marking is to facilitate the training of the tooth recognition and segmentation model. After the tooth recognition and segmentation model is trained, there is no need to mark the panoramic dental radiographs anymore, and the tooth recognition and segmentation model can be directly used to quickly judge the alveolar bone resorption of each tooth on each new panoramic dental radiograph

[0039] The first SegFormer model of this embodiment includes two Transformer encoding modules. The steps of using the first SegFormer model to segment the tooth regions of all panoramic dental radiographs in step S12 include the following steps: taking the circumscribed rectangle of the first alveolar bone resorption dataset label as the tooth region label to obtain the tooth region dataset, and using the first SegFormer model including two Transformer encoding modules to train the tooth region dataset to segment the tooth regions in the panoramic dental radiographs. The second SegFormer model of this embodiment includes four Transformer encoding modules, that is, the SegFormer model for segmenting alveolar bone resorption includes four Transformer encoding modules. The data augmentation method in step S13 of this embodiment includes horizontal flipping, vertical flipping, and contrast transformation

[0040] In this embodiment, the tooth regions in the panoramic dental radiographs are obtained through the first SegFormer model, and then the second SegFormer model is used to continue segmenting the alveolar bone resorption in the tooth regions of the panoramic dental radiographs. During training, first train the first SegFormer model to only segment the tooth regions in the panoramic dental radiographs, and then train the second SegFormer model to continue the bone resorption segmentation on the results segmented by the first SegFormer model

[0041] The alveolar bone resorption label includes teeth except wisdom teeth. The circumscribed rectangle is the smallest circumscribed rectangle that can just include these teeth. The four points of the circumscribed rectangle are determined by judging the minimum and maximum coordinates of the rows and columns of the bone resorption label. The four coordinate points are (minimum row, minimum column), (minimum row, maximum column), (maximum row, minimum column), (maximum row, maximum column)

[0042] The SegFormer model consists of an encoder and a decoder. The role of the encoder is feature extraction. It uses the Transformer encoding module, and through multiple Transformer encoding modules, multi-scale features are obtained by downsampling. In the decoder stage, the multi-scale features are fused through the MLP, and the encoded information is restored to the input size to achieve segmentation. In the encoder stage, generally, the more Transformer encoding modules are used, the richer multi-scale information can be obtained, and a better segmentation effect can be achieved. However, it has higher requirements for hardware and is prone to overfitting. Therefore, the first SegFormer model in this embodiment uses two connected Transformer encoding modules for feature extraction. After the tooth area is segmented, the hardware requirements are reduced. So, the second SegFormer model in this embodiment uses four Transformer encoding modules for feature extraction, which can greatly improve the segmentation accuracy of alveolar bone resorption.

[0043] The panoramic dental film has two characteristics: First, the size of the panoramic dental film image is relatively large, around 3000×1500. If the deep learning method is used, it has relatively high requirements for video memory. In addition, since the supraosseous part of the tooth root is a small target, if the image is directly scaled to reduce the size of the neural network input image, it will lead to resolution loss and the segmentation accuracy will be greatly reduced. Second, the data volume of the panoramic dental film is small, and when taking the panoramic dental film, not only the images of the teeth can be obtained, but also various other oral and maxillofacial tissue structures can be photographed, such as part of the nasal cavity, orbit, temporomandibular joint, hyoid bone, etc. The shapes are complex and the pixel intensities are different, which will interfere with the recognition of teeth. The third molar is located at both ends of the dental arch and is considered a non-functional tooth. It is usually not considered in the diagnosis of periodontitis, and the third molar often shows situations such as inclination and intraosseous impaction. With less data volume and more features in a single image, it is very easy to overfit when using the deep learning method, resulting in poor prediction effects.

[0044] As Figure 5 shown, the two-stage alveolar bone resorption based on SegFormer in this embodiment includes the following steps in the model training stage:

[0045] Step 1: First, use the first SegFormer model containing two Transformer encoding modules to segment the tooth area. The task of this step is relatively simple, so a relatively lightweight model is used, only using two Transformer encoding modules, reducing the need for video memory and removing the interference of the remaining areas. In the case of a small amount of data itself, reducing the useless features of the remaining areas helps to improve the generalization ability of the network.

[0046] Step 2: In the model training stage, the Transformer series requires a large amount of data. Therefore, data augmentation is performed on the data containing only the tooth region by means of horizontal flipping, vertical flipping, contrast transformation, etc., to improve the robustness and generalization ability of the model. The dataset is expanded from the alveolar bone resorption dataset containing 760 panoramic dental radiographs and labels to 4,100 images.

[0047] Step 3: After the image cropping in Step 1, the input image in this step is smaller, and a model with higher precision can be used. Therefore, the second SegFormer model containing four Transformer encoding modules is used to segment the three parts of the tooth crown, supra-bony and sub-bony parts of the tooth root, and training is performed on the data after data augmentation to obtain the final segmentation result.

[0048] Panoramic dental radiographs usually have a large resolution. Inputting the complete picture into the neural network model requires high hardware requirements. If the input image size is reduced by scaling to lower the resolution, the alveolar bone resorption will become blurred and more difficult to segment. Especially the supra-bony part of the tooth root is a small area region, and it is difficult to obtain a good segmentation effect after reducing the resolution. In addition, due to the limitation of the receptive field, the segmentation model based on the convolutional neural network is difficult to process panoramic dental radiographs with large resolution, while the Transformer encoding module of SegFormer can easily obtain a large receptive field. In summary, the two-step segmentation method based on SegFormer in this embodiment, that is, first segmenting the tooth region and then segmenting the alveolar bone resorption, can segment the alveolar bone resorption data without reducing the resolution, so as to achieve high-precision segmentation of the alveolar bone resorption in panoramic dental radiographs.

[0049] In this embodiment, data augmentation is used in the model training stage to cope with the situation of less tooth image data. Its main purpose is to expand the size of the training set to prevent overfitting and enhance the generalization ability of the model. It generates new training samples by performing various transformations on the training data, thereby expanding the training set. The data augmentation in this embodiment can achieve the following purposes: (1) Prevent overfitting: Overfitting is a common problem in machine learning, which is manifested as the model performing well on the training set but poorly on the test set. Through data augmentation, the quantity and diversity of training data can be effectively increased, and the risk of overfitting can be reduced. (2) Enhance the generalization ability of the model: By transforming the training data, the model can learn more data distributions and features, which helps to improve the prediction performance of the model for new data. (3) Solve the problem of insufficient data: In many cases, collecting and annotating a large amount of training data is an expensive and time-consuming task. Through data augmentation, more training data can be generated from limited training samples.

[0050] In this embodiment, the percentage of alveolar bone resorption in the root length can be obtained by calculating based on the distances from the cementoenamel junction (CEJ) to the alveolar bone crest and from the CEJ to the root apex in the segmentation results, and a preliminary diagnosis of periodontitis can be further made. Each panoramic tomogram contains all the teeth of a person's entire mouth, usually 28 - 32 teeth. When making a judgment, the alveolar bone resorption of each tooth in each panoramic tomogram is judged independently, and then the situation of alveolar bone resorption in the entire mouth can be obtained.

[0051] The specific embodiments described above further elaborate on the object, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only the specific embodiments of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An alveolar bone resorption judgment method based on SegFormer and dental panoramic tomography, characterized in that It includes a training step for a tooth recognition and segmentation model for training a tooth recognition and segmentation model, and a step for judging the alveolar bone resorption of each tooth in an orthopantomogram based on the tooth recognition and segmentation model. The tooth recognition and segmentation model includes a first SegFormer model and a second SegFormer model. The first SegFormer model is used to segment the tooth area in the orthopantomogram, and the second SegFormer model is used to segment the crown, supra-bony part of the root, and sub-bony part of the root of the tooth in the orthopantomogram that only contains the tooth area. The training step of the tooth recognition and segmentation model includes the following steps: Step S11: Obtain multiple orthopantomograms and mark the tooth areas of each orthopantomogram by means of color filling, and use all the marked orthopantomograms as the first alveolar bone resorption data to form a first alveolar bone resorption data set. Among them, the tooth area includes the crown, supra-bony part of the root, and sub-bony part of the root, and different colors are used for filling the crown, supra-bony part of the root, and sub-bony part of the root. Step S12: Use the first SegFormer model to segment the tooth areas of all orthopantomograms, and remove the non-tooth areas in the orthopantomograms to obtain orthopantomograms that only contain the tooth area as the second alveolar bone resorption data to form a second alveolar bone resorption data set. Step S13: Perform data augmentation on the second alveolar bone resorption data in the second alveolar bone resorption data set. Step S14: Use the second SegFormer model to segment the crown, supra-bony part of the root, and sub-bony part of the root of the tooth in the second alveolar bone resorption data after data augmentation to obtain the segmentation result. The step for judging the alveolar bone resorption includes the following steps: Step S21: Use the first SegFormer model to segment the tooth area of the orthopantomogram to be judged for alveolar bone resorption, and remove the non-tooth areas. Step S22: Use the second SegFormer model to segment the crown, supra-bony part of the root, and sub-bony part of the root of the tooth in the tooth area to obtain the segmentation result. Step S23: Calculate the percentage of alveolar bone resorption of each tooth in the orthopantomogram accounting for the root length according to the segmentation result to obtain the alveolar bone resorption judgment result of each tooth.

2. The alveolar bone resorption judgment method based on SegFormer and panoramic dental tomograms according to claim 1, wherein When marking in step S11, the enamel-cementum junction points at the mesial and distal sides of the tooth are used as the demarcation points between the crown and the root, and the alveolar bone level points at the mesial and distal sides of the tooth are used as the demarcation points between the supra-bony part of the root and the sub-bony part of the root.

3. The alveolar bone resorption judgment method based on SegFormer and panoramic dental radiographs according to claim 1, wherein, When marking in step S11, the wisdom teeth on the orthopantomogram and the teeth that have lost the enamel-cementum junction points are not marked; when it is impossible to judge the enamel-cementum junction point due to tooth repair or filling treatment, the root-side edge of the restoration or filling on the tooth contour is used as the enamel-cementum junction point.

4. The alveolar bone resorption judgment method based on SegFormer and dental panoramic tomography according to claim 1, wherein The first SegFormer model includes two Transformer encoding modules. The step of using the first SegFormer model to segment the tooth areas of all orthopantomograms in step S12 includes the following steps: The circumscribed rectangle of the first alveolar bone resorption dataset label is used as the tooth region label to obtain the tooth region dataset, and the first SegFormer model including two Transformer encoding modules is used to train the tooth region dataset to segment the tooth region in the orthopantomogram.

5. The alveolar bone resorption judgment method based on SegFormer and dental panoramic tomogram according to claim 1, wherein, The data augmentation method in the step S13 includes horizontal flipping, vertical flipping and contrast transformation.

6. The alveolar bone resorption judgment method based on SegFormer and dental panoramic tomogram according to claim 1, wherein The second SegFormer model includes four Transformer encoding modules.

7. The alveolar bone resorption judgment method based on SegFormer and panoramic dental films according to any one of claims 1 to 6, characterized in that, The percentage of alveolar bone resorption in the root length in the step S23 is obtained by calculating the ratio of the distance between the cementoenamel junction and the alveolar bone crest top to the distance between the cementoenamel junction and the apex point according to the segmentation result.

Images