CN-121999220-A - Tooth instance segmentation method, system, electronic equipment and storage medium

CN121999220ACN 121999220 ACN121999220 ACN 121999220ACN-121999220-A

Abstract

The application discloses a tooth example segmentation method, a system, electronic equipment and a storage medium, wherein the method comprises the steps of obtaining CBCT three-dimensional volume data of a patient, and carrying out segmentation processing on the CBCT three-dimensional volume data to obtain initial segmentation mask data; the method comprises the steps of splicing CBCT three-dimensional volume data and initial segmentation mask data to obtain fusion tensor, inputting the fusion tensor into a segmentation network, outputting a tooth category probability feature map, analyzing the tooth category probability feature map through a maximum value index and a connected domain to obtain each tooth instance and corresponding FDI numbers, fitting and generating dental arch curves according to each tooth instance, obtaining missing FDI numbers according to the FDI numbers, and obtaining central coordinates of dental defect sites according to dental arch curves of adjacent teeth corresponding to the missing FDI numbers. Through the segmentation network, the problems of over-segmentation, under-segmentation and number mismatch are effectively suppressed, and each tooth is ensured to be precisely segmented in a complete anatomical entity form and is assigned with a correct FDI number.

Inventors

WANG SUYU
ZHOU YUFEI
LI XIAOGUANG
ZHANG YU
ZHU NING
LU JUNXI
LIANG SHUYING

Assignees

北京工业大学
北京大学口腔医学院

Dates

Publication Date: 20260508
Application Date: 20260122

Claims (9)

1. A method of tooth instance segmentation, comprising: Acquiring CBCT three-dimensional volume data of a patient, and performing segmentation processing on the CBCT three-dimensional volume data to obtain initial segmentation mask data; splicing the CBCT three-dimensional volume data and the initial segmentation mask data to obtain a fusion tensor; Inputting the fusion tensor into a segmentation network, and outputting a tooth class probability feature map, wherein a loss function in the segmentation network is as follows Wherein L WCE is a weighted cross entropy loss, L FDice is a foreground Dice loss, and alpha and beta are weight parameters; the tooth category probability feature map obtains each tooth instance and the corresponding FDI number through maximum value index and connected domain analysis; Fitting to generate a dental arch curve according to each tooth example, and obtaining a missing FDI number according to the FDI number; and obtaining the central coordinates of the missing tooth site according to the dental arch curves of the adjacent teeth corresponding to the missing FDI numbers.
2. The method for tooth example segmentation according to claim 1, wherein the performing segmentation processing on the CBCT three-dimensional volume data to obtain initial segmentation mask data comprises: And performing segmentation processing on the CBCT three-dimensional data by using an initial segmentation network to obtain initial segmentation mask data, wherein the initial tooth segmentation mask data comprises the shape, the spatial position and the initial FDI label information of each tooth.
3. The tooth example segmentation method according to claim 2, characterized in that the initial tooth segmentation mask is subjected to connected domain analysis and geometric feature calculation to detect over-segmented regions of interest and potentially cross-tooth merged regions of interest; Calculating the length and the volume of a main shaft of each connected domain along the direction of a dental arch fitting curve, and marking the connected domain and the neighborhood thereof as the potential dental arch merging interested region if the length or the volume of the main shaft is larger than 1.5 times of the upper limit of the normal anatomical dimension of a single tooth; The detection mode of the excessive region of interest is that each unique non-zero FDI number is used for judging whether the corresponding voxel forms a single connected region, if a plurality of non-connected regions are formed, each non-connected region and a cube neighborhood with the centroid as the center and the side length of 32 voxels are marked as the excessive region of interest.
4. The method of claim 1, wherein generating a dental arch curve from each of the dental instances by fitting, and obtaining a missing FDI number from the FDI numbers, comprises: According to the tooth examples, centroid coordinates of each tooth example are obtained, and personalized dental arch curves are generated through three times of B spline curve fitting by utilizing all the centroid coordinates and the positions of the centroid coordinates on a jaw bone; according to the FDI numbering system, a complete continuous standard FDI numbering sequence is generated, the actually detected tooth examples are mapped to the standard FDI numbering sequence according to the FDI numbers, and continuously missing FDI numbering segments in the sequence are identified as tooth missing areas.
5. The method according to claim 1, wherein the obtaining the center coordinates of the missing tooth site from the arch curves of the adjacent teeth corresponding to the missing FDI numbers includes: According to bow curve parameters corresponding to adjacent teeth in the standard FDI number sequence of the missing tooth region, obtaining curve parameters corresponding to the missing tooth region through linear interpolation or anatomical average size of missing tooth positions, substituting the curve parameters into an equation of the dental bow curve, and calculating to obtain center coordinates of the missing tooth positions.
6. The method according to claim 5, wherein the obtaining the center coordinates of the missing tooth sites according to the dental arch curves of the adjacent teeth corresponding to the missing FDI numbers further comprises correcting the interpolated curve parameters according to the anatomical form parameters of the teeth around the missing tooth regions so that the center coordinates of the missing tooth sites meet the clinical implant anatomical requirements.
7. A dental instance segmentation system, comprising: The acquisition module is used for acquiring CBCT three-dimensional volume data of a patient, and dividing the CBCT three-dimensional volume data to obtain initial division mask data; The processing module is used for splicing the CBCT three-dimensional data and the initial segmentation mask data to obtain a fusion tensor, inputting the fusion tensor into a segmentation network to output a tooth class probability feature map, wherein a loss function in the segmentation network is as follows Wherein L WCE is weighted cross entropy loss, L FDice is foreground price loss, alpha and beta are weight parameters, the tooth class probability feature map obtains each tooth instance and corresponding FDI number thereof through maximum value index and connected domain analysis, a dental arch curve is generated by fitting according to each tooth instance, and the missing FDI number is obtained according to the FDI number; And the execution module is used for obtaining the central coordinate of the missing tooth site according to the dental arch curve of the adjacent teeth corresponding to the missing FDI number.
8. An electronic device comprising at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program that, when executed by the processing unit, causes the processing unit to perform the method of any of claims 1-6.
9. A storage medium storing a computer program executable by an electronic device, the program when run on the electronic device causing the electronic device to perform the method of any one of claims 1-6.

Description

Tooth instance segmentation method, system, electronic equipment and storage medium Technical Field The present application relates to the field of image processing technologies, and in particular, to a method and a system for segmenting a tooth instance, an electronic device, and a storage medium. Background Cone beam computed tomography (CBCT, cone Beam Computed Tomography) has become a core technical means for visualization of the oromaxillofacial anatomy by virtue of its technical advantages of cost controllability, low radiation dose, and high spatial resolution. Tooth example segmentation (namely, precisely separating single teeth and distributing standard FDI numbers) and missing tooth detection are basic preconditions of personalized operation planning of the oral cavity, and the result accuracy directly influences the rationality and operation safety of the implant implantation position. Currently, related technologies for CBCT image tooth segmentation have been widely explored, and the mainstream schemes include deep learning methods based on 3D U-Net, V-Net, nnUNet, swinUNETR, and other architectures. However, the prior art has significant limitations that a partial scheme does not perform instance-level segmentation and standardized FDI number assignment on existing teeth, and even if the partial scheme supports segmentation and numbering, the problems of over-segmentation (a single tooth is split into a plurality of spatially discontinuous fragments), under-segmentation (a plurality of adjacent teeth are combined by mistake), and FDI number mismatch still generally exist as a result, so that the dentition instance count is deviated. The accurate missing tooth positioning is based on complete checking of existing teeth and correct FDI number sequence analysis, and the technical defects can directly cause the problem that missing tooth facts are covered or missing tooth gaps are misjudged, so that missing tooth detection results are insufficient in reliability, and the application requirements of clinical personalized planting planning on high-precision structured data cannot be met. Disclosure of Invention In view of the above-mentioned shortcomings, the present application provides a method, system, electronic device and storage medium for tooth instance segmentation. In a first aspect, to achieve the above object, the present application provides a tooth example segmentation method, including: Acquiring CBCT three-dimensional volume data of a patient, and performing segmentation processing on the CBCT three-dimensional volume data to obtain initial segmentation mask data; Splicing the CBCT three-dimensional volume data and the initial segmentation mask data to obtain a fusion tensor; Inputting the fusion tensor into a segmentation network, outputting a tooth class probability feature map, and obtaining a loss function in the segmentation network as Wherein L WCE is a weighted cross entropy loss, L FDice is a foreground Dice loss, and alpha and beta are weight parameters; The tooth class probability feature map obtains each tooth instance and the corresponding FDI number through maximum value index and connected domain analysis; Fitting to generate a dental arch curve according to each tooth example, and obtaining a missing FDI number according to the FDI number; And obtaining the central coordinates of the missing tooth sites according to the dental arch curves of the adjacent teeth corresponding to the missing FDI numbers. Further, the method for performing segmentation processing on the CBCT three-dimensional volume data to obtain initial segmentation mask data comprises the following steps: And performing segmentation processing on the CBCT three-dimensional data by using an initial segmentation network to obtain initial segmentation mask data, wherein the initial tooth segmentation mask data comprises the shape, the spatial position and the preliminary FDI label information of each tooth. Further, carrying out connected domain analysis and geometric feature calculation on the initial tooth segmentation mask, and detecting an over-segmentation region of interest and a potential cross-tooth merging region of interest; Calculating the length and the volume of a main shaft of each connected domain along the fitting curve direction of the dental arch, and marking the connected domain and the neighborhood thereof as the potential tooth-crossing merging interested region if the length or the volume of the main shaft is larger than 1.5 times of the upper limit of the normal anatomical dimension of a single tooth; The detection mode of the over-segmentation interested region is that each unique non-zero FDI number is used for judging whether the corresponding voxel forms a single connected region, if a plurality of non-connected regions are formed, each non-connected region and a cube neighborhood with the centroid as the center and the side length of 32 voxels are marked as the over-segmentation i