WO2023246462A1

WO2023246462A1 - Method for segmenting permanent teeth and deciduous teeth and determining tooth numbers in panoramic dental radiograph

Info

Publication number: WO2023246462A1
Application number: PCT/CN2023/097791
Authority: WO
Inventors: 马成龙; 丁王辉; 金阳春; 赵泽宇
Original assignee: 杭州朝厚信息科技有限公司
Priority date: 2022-06-20
Filing date: 2023-06-01
Publication date: 2023-12-28
Also published as: CN117788355A

Abstract

Provided in one aspect of the present application is a method for segmenting permanent teeth and deciduous teeth and determining tooth numbers in a panoramic dental radiograph, which is executed by a computer. The method comprises: acquiring a panoramic dental radiograph; segmenting and classifying teeth in the panoramic dental radiograph by using a trained first deep neural network, wherein the classification is to classify permanent teeth according to respective corresponding tooth numbers, and divide all deciduous teeth into the same class; and for each deciduous tooth, determining the tooth number thereof on the basis of the tooth number of an adjacent permanent tooth in the same tooth jaw.

Description

Methods for segmenting permanent teeth and deciduous teeth and determining tooth numbers in oral panoramic radiographs

Technical field

The present application generally relates to a method for segmenting permanent teeth and deciduous teeth and determining tooth numbers in oral panoramic radiographs.

Background technique

Today, dental care increasingly relies on computer technology to improve the efficiency and accuracy of diagnosis and treatment.

In dental diagnosis and treatment, dental professionals often use oral panoramic radiographs. Oral panoramic radiographs are usually taken outside the mouth using

Oral panoramic radiographs can reflect abnormal phenomena such as missing and damaged teeth, root canal fillings, and bone islands. This information can be used to evaluate the patient's tooth growth and oral health status, and help dental professionals formulate treatment plans.

Identifying permanent teeth and deciduous teeth in oral panoramic radiographs is the basis for intelligent analysis of oral panoramic radiographs. It can not only judge the patient's dental health based on this, such as whether there are missing teeth and the development status of deciduous teeth, but also provides a basis for further intelligent analysis. , for example, determining the tooth number of caries and determining impacted teeth.

Therefore, it is necessary to provide a computer-implemented method for segmentation and tooth number determination of permanent teeth and deciduous teeth in oral panoramic radiographs.

Contents of the invention

One aspect of the present application provides a computer-implemented method for segmenting permanent teeth and deciduous teeth and determining tooth numbers in oral panoramic films, which includes: obtaining an oral panoramic radiograph; using a trained first deep neural network to analyze the oral panoramic The teeth in the film are segmented and classified, wherein the classification is to classify the permanent teeth according to their corresponding Classify the tooth numbers of all deciduous teeth into the same category; and for each deciduous tooth, determine its tooth number based on the tooth number of the adjacent permanent teeth in the same jaw.

In some embodiments, the computer-implemented method for segmenting permanent teeth and deciduous teeth and determining tooth numbers in the oral panoramic photo also includes: using a trained second deep neural network to crop out tooth regions from the oral panoramic photo. Image, the segmentation and classification are performed based on the tooth region image.

In some embodiments, the second deep neural network is yolov3.

In some embodiments, the computer-executed method for segmenting permanent teeth and deciduous teeth and determining tooth numbers in oral panoramic films also includes: using a trained third deep neural network to detect the direction of the tooth region image; and if If the direction of the tooth region image is inconsistent with a predetermined direction, the direction of the tooth region image is corrected, and the segmentation and classification are performed based on the direction corrected tooth region image.

In some implementations, the third deep neural network is ResNet.

In some embodiments, the computer-implemented method for segmenting permanent teeth and deciduous teeth and determining tooth numbers in oral panoramic films also includes: for each deciduous tooth, find three adjacent permanent teeth in the same jaw; and determining the tooth number of the deciduous tooth based on the tooth number of the permanent tooth whose central axis is closest to the center of mass of the deciduous tooth.

In some embodiments, the three permanent teeth adjacent to the deciduous tooth are the three permanent teeth whose center of mass is closest to the center of mass of the deciduous tooth.

In some embodiments, the central axis of the permanent teeth is calculated using principal component analysis.

In some implementations, the first deep neural network is a Mask R-CNN network.

Description of the drawings

The above and other features of the present application will be further described below in conjunction with the accompanying drawings and the detailed description thereof. It should be understood that these drawings only illustrate several exemplary embodiments according to the present application and therefore should not be regarded as limiting the scope of the present application. Unless otherwise indicated, the drawings are not necessarily to scale, and Similar reference numbers indicate similar components.

Figure 1 is a schematic flow chart of a method for segmenting permanent teeth and deciduous teeth and determining tooth numbers in an oral panoramic radiograph executed by a computer in one embodiment of the present application;

Figure 2A is an example of an oral panoramic radiograph showing an interface of a computer program for segmenting permanent teeth and deciduous teeth and determining tooth numbers in an oral panoramic radiograph according to an embodiment of the present application;

Figure 2B is an image of the tooth region cropped from the oral panorama shown in Figure 2A by the target detection network displayed on an interface of the computer program;

Figure 2C is an interface of the computer program showing the tooth area image shown in Figure 2B obtained after the direction is corrected by the direction detection network; and

Figure 3 is an example of the relationship between a deciduous tooth and three adjacent permanent teeth shown in an interface of the computer program.

Detailed ways

The following detailed description refers to the accompanying drawings which form a part of this specification. The illustrative embodiments mentioned in the description and drawings are for illustrative purposes only and are not intended to limit the scope of the present application. Inspired by this application, those skilled in the art will understand that many other embodiments may be adopted, and various changes may be made to the described embodiments without departing from the gist and protection scope of this application. It should be understood that the various aspects of the present application described and illustrated herein can be arranged, replaced, combined, separated and designed in many different configurations, and that these different configurations are within the scope of the present application.

One aspect of the present application provides a computer-implemented method for segmenting permanent teeth and deciduous teeth and determining tooth numbers in oral panoramic radiographs.

Please refer to FIG. 1 , which is a schematic flow chart of a computer-executed method 100 for segmenting permanent teeth and deciduous teeth and determining tooth numbers in an oral panoramic radiograph in one embodiment of the present application.

In one embodiment, the computer-executed method 100 for segmenting permanent teeth and deciduous teeth and determining tooth numbers in oral panoramic radiographs is executed by a computer. Accordingly, another aspect of the present application provides a computer system for segmenting permanent teeth and deciduous teeth in oral panoramic films and determining tooth numbers, which includes a storage device and a processor, and the storage device stores a computer program. , when executed by the processor, the method 100 for segmenting permanent teeth and deciduous teeth and determining tooth numbers in the oral panoramic radiograph will be executed.

In 101, the oral panoramic radiograph to be processed is obtained.

Usually, oral panoramic pictures can be taken using a dental panoramic X-ray machine.

The method 100 of the present application for segmenting permanent teeth and deciduous teeth and determining tooth numbers in oral panoramic pictures can not only process original oral panoramic pictures, but also can process duplicated oral panoramic pictures, for example, using a mobile phone to remake paper oral panoramic pictures. Get the picture.

Please refer to FIG. 2A , which shows an example of an interface of a computer program for segmenting permanent teeth and deciduous teeth and determining tooth numbers in an oral panoramic photograph according to an embodiment of the present application.

In 103, a tooth region image is cropped from the oral panorama.

In one embodiment, a deep neural network can be used to segment the teeth in the oral panoramic photo and determine the tooth number. Hereinafter, the deep neural network is called a segmentation network. Since the teeth are only located in part of the oral panorama, this area will be called the tooth area below. In order to facilitate the training of the tooth segmentation network and improve the segmentation accuracy, the tooth area can be cropped from the oral panorama. Based on the cropped teeth Regional images for tooth segmentation.

In one embodiment, a trained deep neural network can be used to detect the tooth region in the oral panorama. The deep neural network is hereinafter referred to as an object detection network. In one embodiment, the target detection network can use the yolov3 network. It can be understood that in addition to the yolov3 network, the target detection network can also use any other applicable deep neural network, such as Faster RCNN, etc.

Please refer to FIG. 2B , which shows an image of the tooth region cropped from the oral panorama shown in FIG. 2A by the target detection network displayed on an interface of the computer program.

It can be understood that although tooth segmentation based on tooth region images has advantages in network training and segmentation accuracy, it is also feasible to directly segment teeth based on the original oral panorama. Therefore, in another embodiment, this solution can be used .

In 105, the direction of the tooth region image is detected and corrected.

In one embodiment, the segmentation network can be trained with a tooth region image in a predetermined orientation (for example, the maxillary teeth are above and the mandibular teeth are below). Then, the trained segmentation network is used to segment the tooth region image. In order to ensure the accuracy of segmentation and classification, the direction of the input tooth region image can be corrected to make it consistent with the predetermined direction.

In one embodiment, a trained deep neural network can be used to detect the direction of the tooth region image, and correct the tooth region image to the predetermined orientation based on the detection result. The deep neural network is hereinafter referred to as a direction detection network. In one embodiment, the direction detection network can use a ResNet network. It can be understood that in addition to the ResNet network, the direction detection network can also use any other applicable deep neural network, such as EfficientNet, etc.

In one embodiment, the predetermined orientation may be that the maxillary teeth are located above the mandibular teeth. The direction of the tooth area image shown in Figure 2B is that the maxillary teeth are located to the right of the mandibular teeth. It needs to be rotated 90 degrees in the counterclockwise direction to make the It is consistent with the predetermined orientation.

Please refer to FIG. 2C , which shows an interface of the computer program showing the tooth region image shown in FIG. 2B after the direction is adjusted by the direction detection network.

Under the inspiration of this application, it can be understood that the order of clipping and direction correction of the dental region image can be interchanged, that is, first perform direction detection and correction on the oral panoramic picture, and then crop out the corrected direction from the oral panoramic picture. Image of the oral cavity area.

In 107, the corrected-oriented tooth region image is segmented and classified.

In one embodiment, a trained deep neural network can be used to segment and classify teeth in the tooth region image, which is hereinafter referred to as a segmentation network. In one embodiment, the segmentation network can use Mask R-CNN network. It can be understood that in addition to Mask R-CNN network, The segmentation network can also adopt any other suitable deep neural network, such as PointRend, etc.

Teeth include permanent teeth and deciduous teeth, of which there are 32 permanent teeth and 20 deciduous teeth. In one embodiment, permanent and deciduous teeth may be numbered as shown in Table 1. Due to the complexity of the development of deciduous teeth, it is difficult to directly determine the tooth number using a deep neural network. Considering the corresponding relationship between deciduous teeth and permanent teeth, in one embodiment, the deciduous tooth tooth number can be inferred based on the permanent tooth number adjacent to the deciduous tooth. For example, if a deciduous tooth is closest to permanent tooth 11, it can be deduced from Table 1 below that the deciduous tooth number is 51.

Table 1

Correspondingly, the segmentation network classifies the detected teeth into 33 categories, where 32 categories correspond to 32 permanent tooth numbers, and all deciduous teeth are regarded as one category. Segmentation is essentially segmentation while detecting the position of the teeth and determining the category of the teeth. This can be called instance segmentation. The segmentation network outputs all detected teeth and their corresponding segmentation maps and categories.

In 109, the tooth number of the deciduous tooth is determined based on the tooth number of the permanent tooth adjacent to the deciduous tooth.

In one embodiment, the deciduous tooth size can be determined by the following method.

First, for a baby tooth, find the three closest permanent teeth. In one embodiment, the distance between the center of mass of the deciduous tooth and the centroid of all permanent teeth can be calculated, and based on this, three adjacent permanent teeth (permanent teeth belonging to the same jaw as the deciduous tooth) are selected.

Next, calculate the distance between the center of mass of the deciduous tooth and the central axis of the three adjacent permanent teeth, and divide the central axis into The permanent tooth whose axis is closest to the mass center of the deciduous tooth is used as the permanent tooth corresponding to the deciduous tooth, and the tooth number of the deciduous tooth is determined based on the tooth number of the permanent tooth. In one embodiment, a principal component analysis method may be used to calculate the central axis of the tooth.

Please refer to FIG. 3 , which is an example of the relationship between a deciduous tooth and three adjacent permanent teeth shown in an interface of the computer program.

Although various aspects and embodiments of the present application are disclosed herein, other aspects and embodiments of the present application will be apparent to those skilled in the art in light of this application. The various aspects and embodiments disclosed herein are for illustrative purposes only and not for purposes of limitation. The scope and spirit of the present application are determined only by the appended claims.

Likewise, various diagrams may illustrate exemplary architectures or other configurations of the disclosed methods and systems, which may be helpful in understanding the features and functionality that may be included in the disclosed methods and systems. The claims are not limited to the exemplary architectures or configurations shown, as the desirable features may be implemented with various alternative architectures and configurations. Additionally, for flowcharts, functional descriptions, and method claims, the order of blocks presented herein should not be limited to various embodiments implemented in the same order to perform the recited functions, unless the context clearly dictates otherwise. .

Unless expressly stated otherwise, the terms and phrases used herein and variations thereof are to be construed as open-ended and not restrictive. In some instances, the appearance of expansive words and phrases such as "one or more,""atleast,""but not limited to," or other similar language should not be construed as a context in which such expansive language may not be present. The intention or need indicates a narrowing situation.

Claims

A computer-executed method for segmenting permanent teeth and deciduous teeth and determining tooth numbers in oral panoramic films, which includes:

Obtain oral panoramic radiographs;

The trained first deep neural network is used to segment and classify the teeth in the oral panorama, wherein the classification is to classify permanent teeth according to their corresponding tooth numbers and classify all deciduous teeth into the same category; as well as

For each deciduous tooth, its tooth number is determined based on the number of the adjacent permanent tooth in the same jaw.
The computer-implemented method for segmenting permanent teeth and deciduous teeth and determining tooth numbers in oral panoramic films as claimed in claim 1, characterized in that it also includes: using a trained second deep neural network to extract the information from the oral panoramic films. A tooth region image is cropped, and the segmentation and classification are performed based on the tooth region image.
The computer-executed method for segmenting permanent teeth and deciduous teeth and determining tooth numbers in oral panoramic films according to claim 2, characterized in that the second deep neural network is yolov3.
The computer-executed method for segmenting permanent teeth and deciduous teeth and determining tooth numbers in oral panoramic films as claimed in claim 2, characterized in that it further includes:

detecting the direction of the tooth region image using a third trained deep neural network; and

If the direction of the tooth area image is inconsistent with the predetermined direction, the direction of the tooth area image is corrected, and the segmentation and classification are performed based on the direction-corrected tooth area image.
The computer-implemented method for segmenting permanent teeth and deciduous teeth and determining tooth numbers in oral panoramic films according to claim 4, wherein the third deep neural network is ResNet.
The computer-executed method for segmenting permanent teeth and deciduous teeth and determining tooth numbers in oral panoramic films as claimed in claim 1, characterized in that it also includes:

For each deciduous tooth, locate the three adjacent permanent teeth in the same jaw; and

The tooth number of the deciduous tooth is determined based on the tooth number of the permanent tooth whose central axis is closest to the center of mass of the deciduous tooth.
The computer-executed segmentation and tooth number determination method of permanent teeth and deciduous teeth in oral panoramic films as claimed in claim 6, characterized in that the three permanent teeth adjacent to the deciduous teeth have a center of mass equal to the centroid of the deciduous teeth. The three closest permanent teeth.
The computer-executed method for segmenting permanent teeth and deciduous teeth and determining tooth numbers in oral panoramic films as claimed in claim 6, characterized in that the central axis of the permanent teeth is calculated using principal component analysis.
The computer-implemented method for segmenting permanent teeth and deciduous teeth and determining tooth numbers in oral panoramic films as claimed in claim 1, wherein the first deep neural network is a Mask R-CNN network.