CN116823729A - Alveolar bone absorption judging method based on SegFormer and oral cavity curved surface broken sheet - Google Patents

Alveolar bone absorption judging method based on SegFormer and oral cavity curved surface broken sheet Download PDF

Info

Publication number
CN116823729A
CN116823729A CN202310623322.0A CN202310623322A CN116823729A CN 116823729 A CN116823729 A CN 116823729A CN 202310623322 A CN202310623322 A CN 202310623322A CN 116823729 A CN116823729 A CN 116823729A
Authority
CN
China
Prior art keywords
tooth
alveolar bone
curved surface
oral cavity
root
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310623322.0A
Other languages
Chinese (zh)
Inventor
刘程程
于瀚雯
叶鑫
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Sichuan University
Original Assignee
Sichuan University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Sichuan University filed Critical Sichuan University
Priority to CN202310623322.0A priority Critical patent/CN116823729A/en
Publication of CN116823729A publication Critical patent/CN116823729A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G06N3/0455—Auto-encoder networks; Encoder-decoder networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T11/00—2D [Two Dimensional] image generation
    • G06T11/40—Filling a planar surface by adding surface attributes, e.g. colour or texture
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/60—Analysis of geometric attributes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20092—Interactive image processing based on input by user
    • G06T2207/20104—Interactive definition of region of interest [ROI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30036—Dental; Teeth

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Data Mining & Analysis (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Computation (AREA)
  • Databases & Information Systems (AREA)
  • Public Health (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Multimedia (AREA)
  • Software Systems (AREA)
  • Primary Health Care (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Pathology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Epidemiology (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Quality & Reliability (AREA)
  • Geometry (AREA)
  • Radiology & Medical Imaging (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Dental Tools And Instruments Or Auxiliary Dental Instruments (AREA)

Abstract

The application discloses an alveolar bone absorption judging method based on a SegFormer and an oral cavity curved surface fracture slice, which comprises a tooth recognition segmentation model training step and an alveolar bone absorption judging step, wherein the training step comprises the following steps of: obtaining a plurality of mouth cavity curved surface broken layers and marking tooth outlet areas; dividing tooth areas; performing data enhancement on alveolar bone absorption data; the tooth crown, the upper bone root and the lower bone root of the tooth are included in the alveolar bone resorption data after the segmentation data is enhanced. The absorption judging step comprises the following steps: and dividing the dental cavity curved surface fault slice to be judged on the alveolar bone absorption into dental areas by adopting a dental identification and division model, subdividing crowns, upper tooth roots and lower tooth roots of teeth in the dental areas, and then calculating to obtain an alveolar bone absorption judgment result. The application carries out the visual processing of the segmentation result and generates the corresponding alveolar bone absorption measurement value, simplifies the interpretation process of the oral cavity curved surface tomogram, and provides possibility for further operation and generation of more accurate diagnosis.

Description

Alveolar bone absorption judging method based on SegFormer and oral cavity curved surface broken sheet
Technical Field
The application relates to an oral disease diagnosis technology, in particular to an alveolar bone absorption judging method based on SegFormer and an oral curved surface broken sheet.
Background
Periodontitis is one of the most common oral diseases and is the leading cause of loss of teeth in adults. The judgment of alveolar bone resorption level by imaging examination is an important auxiliary examination means for periodontal disease diagnosis. The oral cavity curved surface tomogram is used as an imaging examination means commonly used in the periodontal clinical diagnosis and treatment process, and has the advantages of relatively low price, certain repeatability and the like. When the examination result of the oral cavity curved surface tomography is interpreted, the enamel cementum boundary is taken as the boundary point of the tooth root and the tooth crown, and the height of the alveolar bone on the edge of the near and far middle of the tooth is taken as the level point of the alveolar bone. Based on the diagnostic standard of 2018 international new classification of periodontal disease and peri-implant disease, taking the percentage of the bone resorption of the imaging alveolar to the length of the tooth root and the ratio of the bone resorption to the age as one of the diagnostic standards of periodontitis stage and stage, wherein the percentage of bone resorption is less than 15% in stage I, 15-33% in stage II and > 33% in stage III or IV; the ratio of bone absorption percentage to age is less than 0.25 and is grade A, 0.25-1 is grade B, and more than 1 is grade C. In physiological conditions, the level of alveolar bone is 1-2 mm away from the cementum boundary, and in this range, it is considered that the alveolar bone is not significantly absorbed.
The traditional method for identifying and judging the alveolar bone absorption from the oral cavity curved surface broken layer sheet is manual judgment, subjectivity exists based on the manual judgment result, the quality of the dental caries is limited by the learning level and experience of a reader, and differences exist among different readers. The time required for manually and accurately measuring the alveolar bone absorption degree is long, the feasibility is low in clinical practice, so that an estimation mode is generally used in actual clinical work, however, the estimation is unfavorable for accurate diagnosis at a stage critical value, and the most heavy absorption tooth position cannot be rapidly judged under the condition that the full-mouth alveolar bone absorption is relatively average, so that an accurate diagnosis index is obtained.
Disclosure of Invention
The application aims to solve the problems of strong subjectivity, long time consumption and inaccurate interpretation result in periodontal disease diagnosis of the existing manual interpretation mode, and provides an alveolar bone absorption judgment method based on a segFormer and an oral cavity curved surface broken sheet.
The aim of the application is mainly realized by the following technical scheme:
the alveolar bone absorption judging method based on the SegFormer and the oral cavity curved surface fracture layer comprises a tooth recognition segmentation model training step for training a tooth recognition segmentation model and an alveolar bone absorption judging step for judging the alveolar bone absorption of each tooth in the oral cavity curved surface fracture layer based on the tooth recognition segmentation model, wherein the tooth recognition segmentation model comprises a first SegFormer model and a second SegFormer model, the first SegFormer model is used for segmenting tooth areas in the oral cavity curved surface fracture layer, and the second SegFormer model is used for segmenting crowns, upper tooth roots and lower tooth roots of the teeth in the oral cavity curved surface fracture layer only comprising the tooth areas; the tooth recognition segmentation model training step comprises the following steps of:
s11, acquiring a plurality of oral cavity curved surface fracture pieces, marking tooth areas of the oral cavity curved surface fracture pieces in a color filling mode, and forming a first alveolar bone absorption data set by taking all the marked oral cavity curved surface fracture pieces as first alveolar bone absorption data; the tooth area comprises a crown, an upper bone part tooth root and a lower bone part tooth root, wherein the crown, the upper bone part tooth root and the lower bone part tooth root are filled with different colors;
step S12, dividing the tooth areas of all the oral cavity curved surface fracture slices by adopting a first segFormer model, and removing the non-tooth areas in the oral cavity curved surface fracture slices to obtain the oral cavity curved surface fracture slices only comprising the tooth areas as second alveolar bone absorption data to form a second alveolar bone absorption data set;
step S13, data enhancement is carried out on second alveolar bone absorption data in the second alveolar bone absorption data set;
s14, segmenting the crown, the upper root and the lower root of the tooth in the data of the bone absorption of the second alveolar bone after data enhancement by adopting a second SegFormer model to obtain segmentation results;
the alveolar bone resorption judging step includes the steps of:
s21, dividing a dental region of an oral cavity curved surface fault slice of alveolar bone absorption to be judged by adopting a first SegFormer model, and removing a non-dental region;
s22, segmenting a crown, an upper bone part root and a lower bone part root of a tooth in a tooth area by adopting a second segFormer model to obtain segmentation results;
and S23, calculating the percentage of the alveolar bone absorption of each tooth in the oral cavity curved surface fracture slice to the length of the tooth root according to the segmentation result to obtain the alveolar bone absorption judgment result of each tooth. When the method is applied, the training step of the tooth recognition segmentation model is to improve the training precision, a plurality of oral cavity curved surface broken pieces are adopted, and the alveolar bone absorption judging step is to treat the single oral cavity curved surface broken pieces.
Further, in the step S11, the enamel cementum boundary points of the near middle and far middle teeth are used as boundary points of the crown and the root, and the alveolar bone level points of the near middle and far teeth are used as boundary points of the upper root and the lower root.
Further, the step S11 is to mark the wisdom teeth on the oral cavity curved surface fracture plate and the teeth which have lost the cementum boundary points without marking; when the tooth is subjected to repair or filling treatment and the cementum boundary point cannot be judged, the root edge of the repair body or the filling body on the tooth contour is used as the cementum boundary point.
Further, the first SegFormer model includes two transducer encoding modules, and the step S12 of dividing the tooth regions of all the oral cavity curved surface broken-layer sheets by using the first SegFormer model includes the following steps:
and taking the circumscribed rectangle of the first alveolar bone absorption dataset label as a tooth area label to obtain a tooth area dataset, training the tooth area dataset by adopting a first SegFormer model comprising two transducer coding modules, and dividing the tooth area in the outlet cavity curved surface fracture layer sheet.
Further, the data enhancement method in step S13 includes horizontal flipping, vertical flipping, and contrast transformation.
Further, the second SegFormer model includes four transducer encoding modules.
Further, the percentage of alveolar bone absorption in the root length in step S23 is calculated from the ratio of the distance between the cementum boundary and the crest of the alveolar bone to the distance between the cementum boundary and the root tip according to the segmentation result.
In summary, compared with the prior art, the application has the following beneficial effects:
(1) When the method is applied, firstly, marking three parts of tooth crowns, bone upper part tooth roots and bone lower part tooth roots in the oral cavity curved surface broken layer tablet to manufacture a data set in a model training stage; then, the tooth area in the curved surface broken layer sheet of the outlet cavity is segmented by adopting a first SegFormer model, useless information is removed, and training of the first SegFormer model is achieved; then data enhancement is performed on the alveolar bone resorption data set containing only the tooth region; finally, the second SegFormer model is utilized to segment the crown, the upper tooth root and the lower tooth root of the tooth in the data of the alveolar bone absorption after the data enhancement, so that the training of the second SegFormer model is realized. When the application judges the alveolar bone absorption of each tooth in the alveolar bone fracture slice, a trained first SegFormer model is adopted to divide the dental area of the dental fracture slice of the alveolar bone absorption to be judged, a trained second SegFormer model is adopted to divide the crown, the upper part root and the lower part root of the tooth in the dental area, and then relevant numerical calculation is carried out according to the division result, so that the percentage of the bone absorption to the length of the root is obtained, and the initial diagnosis of periodontitis is further carried out.
(2) When the method is applied, the SegFormer model can realize high-precision segmentation of the alveolar bone absorption of the oral cavity curved surface fault slice, optimize the traditional oral cavity curved surface fault slice diagnosis process and output results, save a great deal of labor cost and reduce the technical sensitivity of periodontal disease diagnosis based on the oral cavity curved surface fault slice.
Drawings
The accompanying drawings, which are included to provide a further understanding of embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings:
FIG. 1 is a flowchart of a tooth recognition segmentation model training step in accordance with one embodiment of the present application;
FIG. 2 is a flowchart showing an alveolar bone resorption determination procedure according to an embodiment of the present application;
FIG. 3 is a schematic illustration of the enamel cementum kingdom in the mesial and distal 25 teeth, and the horizontal point distribution of the sockets in the mesial and distal teeth;
FIG. 4 is a schematic representation of an exemplary embodiment of the present application as applied to an alveolar bone resorption dataset of a curved surface tomogram of an oral cavity;
fig. 5 is a flowchart illustrating a process of a picture corresponding to an alveolar bone absorption dataset during a training phase of a tooth recognition segmentation model according to an embodiment of the present application.
Detailed Description
For the purpose of making apparent the objects, technical solutions and advantages of the present application, the present application will be further described in detail with reference to the following examples and the accompanying drawings, wherein the exemplary embodiments of the present application and the descriptions thereof are for illustrating the present application only and are not to be construed as limiting the present application.
Examples:
as shown in fig. 1 and 2, the alveolar bone absorption determination method based on SegFormer and the oral cavity curved surface fracture layer includes a tooth recognition division model training step for training a tooth recognition division model, and an alveolar bone absorption determination step for determining the alveolar bone absorption of each tooth in the oral cavity curved surface fracture layer based on the tooth recognition division model. The tooth recognition segmentation model of the embodiment comprises a first SegFormer model and a second SegFormer model, wherein the first SegFormer model is used for segmenting tooth areas in the oral cavity curved surface fracture layer sheet, and the second SegFormer model is used for segmenting crowns, upper-bone part tooth roots and lower-bone part tooth roots of teeth in the oral cavity curved surface fracture layer sheet only comprising the tooth areas.
The tooth recognition segmentation model training step of the present embodiment includes the steps of: s11, acquiring a plurality of oral cavity curved surface fracture pieces, marking tooth areas of the oral cavity curved surface fracture pieces in a color filling mode, and forming a first alveolar bone absorption data set by taking all the marked oral cavity curved surface fracture pieces as first alveolar bone absorption data; the tooth area comprises a crown, an upper bone part tooth root and a lower bone part tooth root, wherein the crown, the upper bone part tooth root and the lower bone part tooth root are filled with different colors; step S12, dividing the tooth areas of all the oral cavity curved surface fracture slices by adopting a first segFormer model, and removing the non-tooth areas in the oral cavity curved surface fracture slices to obtain the oral cavity curved surface fracture slices only comprising the tooth areas as second alveolar bone absorption data to form a second alveolar bone absorption data set; step S13, data enhancement is carried out on second alveolar bone absorption data in the second alveolar bone absorption data set; and S14, segmenting the crown, the upper root and the lower root of the tooth in the data of the bone absorption of the second alveolar bone after the data enhancement by adopting a second SegFormer model to obtain a segmentation result.
The alveolar bone resorption judging step of the present embodiment includes the steps of: s21, dividing a dental region of an oral cavity curved surface fault slice of alveolar bone absorption to be judged by adopting a first SegFormer model, and removing a non-dental region; s22, segmenting a crown, an upper bone part root and a lower bone part root of a tooth in a tooth area by adopting a second segFormer model to obtain segmentation results; and S23, calculating the percentage of the alveolar bone absorption of each tooth in the oral cavity curved surface fracture slice to the length of the tooth root according to the segmentation result to obtain the alveolar bone absorption judgment result of each tooth. The ratio of the distance between the cementum boundary and the crest of the alveolar bone to the distance between the cementum boundary and the root tip point is calculated according to the segmentation result to obtain the percentage of the alveolar bone absorption to the root length. The segmentation results obtained in step S14 and step S22 in this embodiment are the crown, the upper root and the lower root.
In this embodiment, the first alveolar bone absorption data of the first alveolar bone absorption dataset is marked by using existing mapping software such as photo hop manually, when marking, the enamel cementum boundary points in the middle and the far of the tooth are used as boundary points between the dental crown and the dental root, the alveolar bone level points in the middle and the far of the tooth are used as boundary points between the dental root of the upper part of the bone and the dental root of the lower part of the bone, namely, the enamel cementum boundary points in the middle and the far of the tooth are used for dividing the dental crown and the dental root, and the enamel cementum level points in the middle and the far of the tooth are used for dividing the upper part of the bone and the lower part of the dental root. The enamel cementum boundary refers to the position where enamel of teeth is adjacent to cementum. The alveolar bone level point refers to the highest point of the alveolar bone crest in the near or far middle of the tooth. As shown in FIG. 3, M1 and D1 represent the cementum kingdom of the mesial and distal 25 teeth, respectively, and M2 and D2 represent the socket level points of the mesial and distal 25 teeth, respectively.
The present embodiment marks the entire tooth according to the contour of the tooth and fills the three portions with different colors, respectively. The judgment of alveolar bone resorption is divided into diagnosis of periodontitis, and the wisdom teeth are not in the category considered by the diagnosis, and in addition, many wisdom teeth are buried in the alveolar bone, so that the judgment of alveolar bone resorption cannot be performed, and therefore, all wisdom teeth are not marked when the embodiment is applied. When the tooth is a stump, the position of the cementum boundary cannot be confirmed, and the stump is usually not reserved for the lost cementum boundary, and the mark is lost, so that the embodiment does not mark when the tooth is a stump and the cementum boundary point is lost. When the tooth is subjected to repair or filling treatment and the cementum boundary point cannot be judged, the root edge of the repair body or the filling body on the tooth contour is used as the cementum boundary point. The tooth with a two-dimensional plane has four directions, namely a crown direction, a root direction, a mesial direction and a distal direction, and the root edge of the restoration or the filling body on the tooth profile is taken as the root edge in the cementum boundary point, namely the edge of the restoration or the filling body close to the tooth root direction.
Fig. 4 shows an example of artificial marking of the alveolar bone absorption data of a curved surface fracture plate of the oral cavity, fig. 4 (a) shows an unlabeled image, and fig. 4 (b) shows a marked image. The purpose of the manual marking is to facilitate the training of the tooth recognition and segmentation model, the tooth recognition and segmentation model is not required to be marked after the training is completed, and the tooth recognition and segmentation model can be directly adopted to rapidly judge the alveolar bone absorption condition of each tooth on each brand-new tooth surface fault.
The first SegFormer model of the present embodiment includes two transducer encoding modules, and the step S12 of dividing the tooth regions of all the oral cavity curved surface broken layers by using the first SegFormer model includes the following steps: and taking the circumscribed rectangle of the first alveolar bone absorption dataset label as a tooth area label to obtain a tooth area dataset, training the tooth area dataset by adopting a first SegFormer model comprising two transducer coding modules, and dividing the tooth area in the outlet cavity curved surface fracture layer sheet. The second SegFormer model of this embodiment includes four transducer coding modules, i.e., the SegFormer model for dividing alveolar bone resorption includes four transducer coding modules. The data enhancement method in step S13 of the present embodiment includes horizontal flipping, vertical flipping, and contrast transformation.
In this embodiment, the tooth region in the oral cavity curved surface fracture layer obtained by the first SegFormer model is then segmented continuously by the second SegFormer model for alveolar bone absorption in the tooth region in the oral cavity curved surface fracture layer. During training, the first SegFormer model is trained to only segment tooth areas in the curved surface fracture plate of the oral cavity, and then the second SegFormer model is trained to continue bone resorption segmentation on the segmentation result of the first segmer model.
The alveolar bone absorption tag includes teeth other than wisdom teeth, the circumscribed rectangle is a right circumscribed rectangle that can just include the teeth, four points of the right circumscribed rectangle are determined by judging the minimum and maximum coordinates of the bone absorption tag row and column, and the four coordinate points are (row minimum, column minimum), (row minimum, column maximum), (row maximum, column minimum), (row maximum, column maximum), respectively.
The SegFormer model comprises an encoder and a decoder, wherein the encoder is used for extracting characteristics, a transform coding module is adopted, and multi-scale characteristics are obtained through downsampling of a plurality of transform coding modules. In the decoder stage, the multi-scale features are fused through the MLP, and the coding information is restored to the input size to realize segmentation. In the encoder stage, more transform encoding modules are adopted to obtain richer multi-scale information generally, so that better segmentation effect is obtained, but the requirements on hardware are higher and fitting is easy. Therefore, the first SegFormer model of the present embodiment adopts two connected transform coding modules to perform feature extraction, and after tooth regions are segmented, the hardware requirement is reduced, so the second SegFormer model of the present embodiment adopts four transform coding modules to perform feature extraction, and can greatly improve alveolar bone absorption segmentation accuracy.
The oral cavity curved surface broken sheet has two characteristics: in addition, since the bone upper part of the tooth root is a small target, if the size of the neural network input image is directly scaled, the resolution loss is caused, and the segmentation precision is greatly reduced. The second point is that the data volume of the oral cavity curved surface fault slice is smaller, and when the oral cavity curved surface fault slice is shot, not only can the image of teeth be obtained, but also other various tissue structures of the oral cavity, jaw and face can be shot, such as partial nasal cavity, eye orbit, temporomandibular joint, hyoid bone and the like, the shape is complex, the pixel intensity is different, and the recognition of the teeth can be interfered. The third molar is located at both ends of the dentition and is considered to be a nonfunctional tooth, which is generally not considered in the diagnosis of periodontitis, and is often subject to tilting, intraosseous burial, and the like. The data volume is small, the characteristics of a single image are more, and the prediction effect is poor due to easy overfitting when a deep learning method is adopted.
As shown in fig. 5, the two-stage alveolar bone resorption based on SegFormer in this embodiment includes the following steps in the model training stage:
step one: the tooth area is segmented by adopting the first SegFormer model comprising two transducer coding modules, the task of the step is simpler, therefore, a lighter model is adopted, only two transducer coding modules are used, the display memory requirement is reduced, the interference of other areas is removed, and the useless characteristics of the other areas are reduced under the condition that the data volume is very small, so that the generalization capability of the network is improved.
Step two: in the model training stage, a large data volume is required by the transducer series, so that data of only the tooth region is enhanced by adopting modes of horizontal overturning, vertical overturning, contrast conversion and the like, the robustness and generalization capability of the model are improved, and the data set is expanded to 4100 pieces from an alveolar bone absorption data set containing 760 pieces of oral cavity curved surface tomograms and labels.
Step three: the image cutting in the first step is smaller in input image, and a model with higher precision can be adopted, so that a second SegFormer model comprising four transducer coding modules is adopted to segment three parts of the crown, the upper bone and the lower bone of the tooth root, and training is carried out on data after data enhancement, so that a final segmentation result is obtained.
The oral cavity curved surface tomogram usually has larger resolution, the input of the complete picture into the neural network model has high requirement on hardware, and if the resolution is reduced by reducing the size of the input image in a zooming mode, the alveolar bone absorption becomes fuzzy and is more difficult to divide, and especially the bone upper part of the tooth root is a small area region, and good dividing effect is difficult to obtain after the resolution is reduced. In addition, the segmentation model based on the convolutional neural network is difficult to process the large-resolution oral cavity curved surface fracture slice due to the limitation of the receptive field, and a transducer coding module of the SegFormer can easily obtain a larger receptive field. In summary, the two-step segmentation method based on SegFormer in this embodiment, that is, the tooth region is segmented first, and the alveolar bone absorption is segmented again, so that the alveolar bone absorption data can be segmented without reducing the resolution, thereby realizing the high-precision segmentation of the alveolar bone absorption of the oral cavity curved surface fault slice.
The data enhancement is used in the model training stage in order to cope with the situation of less tooth image data, and the main purpose of the embodiment is to enlarge the size of the training set to prevent overfitting and enhance the generalization capability of the model. Which expands the training set by generating new training samples by performing various transformations on the training data. The following objects can be achieved by the data enhancement in this embodiment: (1) prevent overfitting: overfitting is a common problem in machine learning, in that models perform well on training sets, but perform poorly on test sets. Through data enhancement, the number and diversity of training data can be effectively increased, and the risk of overfitting is reduced. (2) enhancing model generalization ability: by transforming the training data, the model can learn more data distribution and characteristics, which helps to improve the predictive performance of the model on new data. (3) solve the problem of insufficient data: in many cases, collecting and annotating large amounts of training data is a costly and time consuming task. By data enhancement, more training data can be generated from a limited number of training samples.
According to the method, the percentage of the alveolar bone absorption to the root length can be obtained by calculating according to the alveolar bone boundary-alveolar bone crest distance and the alveolar bone boundary-root point distance of the segmentation result, and the initial diagnosis of periodontitis is further carried out. Each full-mouth curved surface broken layer sheet comprises all teeth of a person, and usually 28-32 teeth. When judging, the alveolar bone absorption of each tooth in each full-mouth curved surface fracture layer is independently judged, and then the full-mouth alveolar bone absorption can be obtained.
The foregoing description of the embodiments has been provided for the purpose of illustrating the general principles of the application, and is not meant to limit the scope of the application, but to limit the application to the particular embodiments, and any modifications, equivalents, improvements, etc. that fall within the spirit and principles of the application are intended to be included within the scope of the application.

Claims (7)

1. The alveolar bone absorption judging method based on the segforce and the oral cavity curved surface broken layer is characterized by comprising a tooth recognition and segmentation model training step for training a tooth recognition and segmentation model and an alveolar bone absorption judging step for judging the alveolar bone absorption of each tooth in the oral cavity curved surface broken layer based on the tooth recognition and segmentation model, wherein the tooth recognition and segmentation model comprises a first segforce model and a second segforce model, the first segforce model is used for dividing tooth areas in the oral cavity curved surface broken layer, and the second segforce model is used for dividing crowns, upper-bone part tooth roots and lower-bone part tooth roots of the teeth in the oral cavity curved surface broken layer which only comprise the tooth areas; the tooth recognition segmentation model training step comprises the following steps of:
s11, acquiring a plurality of oral cavity curved surface fracture pieces, marking tooth areas of the oral cavity curved surface fracture pieces in a color filling mode, and forming a first alveolar bone absorption data set by taking all the marked oral cavity curved surface fracture pieces as first alveolar bone absorption data; the tooth area comprises a crown, an upper bone part tooth root and a lower bone part tooth root, wherein the crown, the upper bone part tooth root and the lower bone part tooth root are filled with different colors;
step S12, dividing the tooth areas of all the oral cavity curved surface fracture slices by adopting a first segFormer model, and removing the non-tooth areas in the oral cavity curved surface fracture slices to obtain the oral cavity curved surface fracture slices only comprising the tooth areas as second alveolar bone absorption data to form a second alveolar bone absorption data set;
step S13, data enhancement is carried out on second alveolar bone absorption data in the second alveolar bone absorption data set;
s14, segmenting the crown, the upper root and the lower root of the tooth in the data of the bone absorption of the second alveolar bone after data enhancement by adopting a second SegFormer model to obtain segmentation results;
the alveolar bone resorption judging step includes the steps of:
s21, dividing a dental region of an oral cavity curved surface fault slice of alveolar bone absorption to be judged by adopting a first SegFormer model, and removing a non-dental region;
s22, segmenting a crown, an upper bone part root and a lower bone part root of a tooth in a tooth area by adopting a second segFormer model to obtain segmentation results;
and S23, calculating the percentage of the alveolar bone absorption of each tooth in the oral cavity curved surface fracture slice to the length of the tooth root according to the segmentation result to obtain the alveolar bone absorption judgment result of each tooth.
2. The method according to claim 1, wherein the step S11 is performed by using the enamel cementum boundary points of the near middle and far middle teeth as the boundary points of the crown and the root, and using the alveolar bone level points of the near middle and far teeth as the boundary points of the upper root and the lower root.
3. The method for determining alveolar bone absorption based on SegFormer and a broken sheet of an oral cavity curved surface according to claim 1, wherein the step S11 is to mark wisdom teeth on the broken sheet of an oral cavity curved surface and teeth that have lost the cementum boundary point without marking; when the tooth is subjected to repair or filling treatment and the cementum boundary point cannot be judged, the root edge of the repair body or the filling body on the tooth contour is used as the cementum boundary point.
4. The method for determining alveolar bone absorption based on SegFormer and oral cavity curved surface broken layer according to claim 1, wherein the first SegFormer model comprises two transducer encoding modules, and the step S12 of dividing the tooth areas of all the oral cavity curved surface broken layer by using the first SegFormer model comprises the following steps:
and taking the circumscribed rectangle of the first alveolar bone absorption dataset label as a tooth area label to obtain a tooth area dataset, training the tooth area dataset by adopting a first SegFormer model comprising two transducer coding modules, and dividing the tooth area in the outlet cavity curved surface fracture layer sheet.
5. The method according to claim 1, wherein the data enhancement method in step S13 comprises horizontal flipping, vertical flipping, and contrast transformation.
6. The method for determining alveolar bone resorption based on SegFormer and oral curved fracture plate according to claim 1, wherein the second SegFormer model comprises four transducer encoding modules.
7. The method according to any one of claims 1 to 6, wherein the percentage of the alveolar bone absorption in the root length in step S23 is obtained by calculating a ratio of a distance between the cementum boundary and the crest of the alveolar bone to a distance between the cementum boundary and the root apex from the segmentation result.
CN202310623322.0A 2023-05-30 2023-05-30 Alveolar bone absorption judging method based on SegFormer and oral cavity curved surface broken sheet Pending CN116823729A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310623322.0A CN116823729A (en) 2023-05-30 2023-05-30 Alveolar bone absorption judging method based on SegFormer and oral cavity curved surface broken sheet

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310623322.0A CN116823729A (en) 2023-05-30 2023-05-30 Alveolar bone absorption judging method based on SegFormer and oral cavity curved surface broken sheet

Publications (1)

Publication Number Publication Date
CN116823729A true CN116823729A (en) 2023-09-29

Family

ID=88128495

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310623322.0A Pending CN116823729A (en) 2023-05-30 2023-05-30 Alveolar bone absorption judging method based on SegFormer and oral cavity curved surface broken sheet

Country Status (1)

Country Link
CN (1) CN116823729A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117455934A (en) * 2023-12-22 2024-01-26 中南大学 Method for enhancing and segmenting abnormal lesion area of oral cavity CT

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117455934A (en) * 2023-12-22 2024-01-26 中南大学 Method for enhancing and segmenting abnormal lesion area of oral cavity CT
CN117455934B (en) * 2023-12-22 2024-03-12 中南大学 Method for enhancing and segmenting abnormal lesion area of oral cavity CT

Similar Documents

Publication Publication Date Title
US11651494B2 (en) Apparatuses and methods for three-dimensional dental segmentation using dental image data
US11734825B2 (en) Segmentation device and method of generating learning model
US20200305808A1 (en) Automated Tooth Localization, Enumeration, and Diagnostic System and Method
US11443423B2 (en) System and method for constructing elements of interest (EoI)-focused panoramas of an oral complex
CN112120810A (en) Three-dimensional data generation method of tooth orthodontic concealed appliance
US20220084267A1 (en) Systems and Methods for Generating Quick-Glance Interactive Diagnostic Reports
CN113223010A (en) Method and system for fully automatically segmenting multiple tissues of oral cavity image
CN113811916A (en) System and method for generating digital three-dimensional tooth models
CN111709959B (en) Digital intelligent diagnosis method for orthodontic treatment
US20210217170A1 (en) System and Method for Classifying a Tooth Condition Based on Landmarked Anthropomorphic Measurements.
CN109767841A (en) A kind of scale model search method and device based on cranium jaw face three-dimensional configuration database
CN114757960B (en) Tooth segmentation and reconstruction method based on CBCT image and storage medium
CN112785609B (en) CBCT tooth segmentation method based on deep learning
EP4144324A1 (en) Intelligent design method for digital model for oral digital impression instrument
CN116823729A (en) Alveolar bone absorption judging method based on SegFormer and oral cavity curved surface broken sheet
CN110236673A (en) Design method and device before a kind of bilateral jaw defect Reconstruction based on database
CN114926470A (en) System and method for segmenting impacted tooth medical image based on deep learning
CN113393470A (en) Full-automatic tooth segmentation method
CN112201349A (en) Orthodontic operation scheme generation system based on artificial intelligence
CN116958169A (en) Tooth segmentation method for three-dimensional dental model
CN115690045A (en) Quantitative assessment method for bone increment before and after periodontitis treatment based on curved surface fault slice
CN115409811A (en) Tooth model reconstruction method, device, equipment and medium based on curvature enhancement
CN112420171B (en) Method and system for classifying maxillary sinus floor bone based on artificial intelligence
CN116883428B (en) Mandible spiral CT image partition segmentation method
CN114681091B (en) Method and equipment for evaluating occlusion condition of dental prosthesis

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination