CN-121981934-A - CBCT measurement system

CN121981934ACN 121981934 ACN121981934 ACN 121981934ACN-121981934-A

Abstract

The invention relates to the technical field of CBCT measurement and discloses a CBCT measurement system which comprises a processor, a data acquisition module, an image preprocessing module, a segmentation model module, an identification module, a parameter measurement module and a result output module, wherein the processor is respectively connected with the image preprocessing module, the segmentation model module, the identification module and the parameter measurement module, the image preprocessing module is also connected with the data acquisition module, the identification module is also connected with the parameter measurement module, and the parameter measurement module is also connected with the result output module. According to the invention, the intelligent measurement of the CBCT image is carried out through the deep learning algorithm, the intelligent segmentation of the relevant anatomical structure of the planting area and the automatic measurement of the relevant parameters of the planting diagnosis are realized, the measurement result is reliable, and the method can be used for the data measurement analysis of the preimplantation planning.

Inventors

LIU XU
Wang Cenguang
LIU XINGYUAN
WANG JIALIN
YANG PI
LIU YUANKAI

Assignees

四川乃康科技有限公司

Dates

Publication Date: 20260505
Application Date: 20240415

Claims (8)

1. The CBCT measurement system is characterized by comprising a processor, a data acquisition module, an image preprocessing module, a segmentation model module, an identification module, a parameter measurement module and a result output module, wherein the processor is respectively connected with the image preprocessing module, the segmentation model module, the identification module and the parameter measurement module, the image preprocessing module is also connected with the data acquisition module, the identification module is also connected with the parameter measurement module, and the parameter measurement module is also connected with the result output module; the data acquisition module acquires CBCT image data by adopting oral cone beam CT scanning equipment; The image preprocessing module is used for preprocessing the CBCT image data acquired by the data acquisition module, wherein the preprocessing comprises resolution normalization and value standardization; the segmentation model module is used for realizing the segmentation of jawbone, maxillary sinus, nose bottom and mandibular nerve tubes by adopting nnU-Net nerve network model; the recognition module is used for recognizing and segmenting teeth by adopting an improved Mask R-CNN neural network model so as to determine tooth positions and tooth missing conditions; the parameter measurement module is used for measuring relevant parameters of planting diagnosis; and the result output module is used for outputting the results of the parameter measurement module and the identification module in an Excel table form and displaying the results at the front end.
2. CBCT measurement system according to claim 1, characterized in that in the identification module the modified Mask R-CNN neural network model comprises 4 phases, in particular: the method comprises the following steps of 1, taking a residual error network ResNet and a feature pyramid network FPN as feature extractors, and extracting features of an input image; Step2, extracting a target region ROI which possibly exists from the obtained feature map through a region generation network RPN; Inputting the ROI obtained in the stage 2 into ROIAlign, and mapping the ROI into a feature vector with a fixed dimension by a bilinear interpolation mode; and 4, respectively inputting the mapped features into 3 branches, carrying out classification and bounding box regression through a full connection layer, and carrying out semantic segmentation through a full convolution layer, wherein the 3 branches are respectively Class score branches, box coordinates branches and improved Masks branches.
3. The CBCT measurement system of claim 2, wherein in the stage 2, the region generating network RPN inputs a feature map of the feature extraction network, and outputs a set of target candidate region rectangular frames; The method comprises the steps of traversing a characteristic diagram output by an FPN through a 3X 3 sliding window, wherein a mapping point of the sliding window center of the current position in an original image pixel space is the anchor point, setting 9 anchor blocks with different sizes (the size of the anchor blocks is 9 in length-width ratio example) by taking the anchor point as the center, obtaining the size and the coordinates of a corresponding area in the original image according to the known anchor point position and the size of the anchor block frame, wherein the area is a preset candidate frame, classifying and frame regression are carried out on 256-dimensional characteristic vectors obtained through convolution through two full-connection layer branches, and then obtaining the area most likely to contain a target through non-maximum suppression.
4. A CBCT measurement system according to claim 3, wherein in stage 3, the ROI alignment uses bilinear interpolation to obtain image values at pixels with floating point coordinates to transform the whole feature aggregation process into a continuous operation; The principle of the method is that 2X 2 pooling operation is carried out on the ROI area, the ROI area is divided into 4 areas with the size of 2X 2, then 4 sampling points and pixel values of 4 characteristic points nearest to the sampling points are selected in each small area, the pixel value of each sampling point is obtained through a bilinear interpolation method, and finally the pooling value of each cell is calculated to generate a characteristic diagram with the size of 2X 2 of the ROI area.
5. The CBCT measurement system of claim 4, wherein in the stage 4, the modified Masks branches perform downsampling encoding by a convolution layer, the deconvolution layer performs upsampling, and shallow high resolution features of different scales are input to the deconvolution layer through a jump connection and SE module while upsampling, and the network structure is specifically as follows: 1. The 2 layers and the 3 layers are all convolution layers, the convolution kernel size is 3, the step length is 1, and each convolution is followed by a batch normalization BN layer and a ReLU activation function; 4. The 5, 6 and 7 layers are deconvolution layers, wherein the convolution kernel size of the 4, 5 and 6 layers is 3, the step length is 1, the convolution kernel size of the 7 th layer is 2, the step length is 2, the 14 x 14 characteristic diagram is input, the 8 x 8 characteristic diagram is obtained through 3 convolution layers, and the 28 x 28 characteristic diagram is obtained through the deconvolution layers; For the input of the deconvolution layer, the symmetrical layers of the encoder network and the decoder network provide jump connection, the result of the convolution operation of each layer of the encoder network is spliced with the result of sampling on the decoder network after passing through the SE module, and finally a binary segmentation mask is generated through the sigmoid layer.
6. The CBCT measurement system of claim 5, wherein the feature fusion manner of the jump connection is a stitching of feature graphs in a channel dimension, and a calculation formula is: Wherein W (h, W, a) and V (h, W, b) are characteristic diagrams from different layers respectively, F (h, W, c) is the characteristic diagram after splicing, h and W are the length and width of the characteristic diagram, and a, b and c are the channel number of the characteristic diagram.
7. The CBCT measurement system of claim 5, wherein a core of the SE module is compression and excitation; firstly, a feature X is convolved to change the channel number from C' to C, a feature map U is transmitted to a compression operation, each feature channel is compressed into a real number by using global average pooling, a receptive field is expanded to a global range, and the compression calculation process is as follows: The method comprises the following steps of obtaining a characteristic diagram, wherein U c is a characteristic diagram obtained after convolution, c is the number of channels of U, H multiplied by W is the space dimension of U, exciting operation captures compressed real number column information, using two full-connection layers to increase nonlinearity of a module, reducing dimension through a first full-connection layer, activating through a rectification linear unit ReLU, increasing dimension through a second full-connection layer, and finally, activating through a sigmoid function, wherein the whole process is as follows: s=σ[W 2 δ(W 1 z)] Wherein, delta is a nonlinear activation function ReLU, W 1 and W 2 are parameters of two full connection layers respectively, sigma is a sigmoid function, finally, the original characteristics are weighted, and channel importance coefficients obtained by excitation operation are multiplied by the original characteristics channel by channel to obtain characteristics with attention information:
8. The CBCT measurement system of claim 1, wherein the parameters measured by the parameter measurement module include a edentulous area mesial-distal repair space, an edentulous area available bone width, an edentulous area available bone height, and an edentulous area mesial-distal root spacing; The near-far middle repairing space of the missing tooth region is characterized in that a point is taken on the far-middle axial surface of the near-middle adjacent tooth, a point is taken on the near-middle axial surface of the far-middle adjacent tooth, and the minimum distance between the two points is measured, which is also called as the near-far middle gap of the missing tooth region; The distance between the bone and the tongue of the alveolar bone of the tooth-missing area with the bone width can be used in the tooth-missing area, the measurement of the parameter data is based on the middle position of the tooth-missing area, the distances between the cheek and tongue of the positions 2, 4, 6, 8, 10, 12 and 14mm below the crest of the tooth-missing are measured, and the cheek and tongue distances of the narrowest part are output as a result; The method comprises the steps of measuring the distance from the apex of an alveolar ridge to each anatomical landmark point, wherein the height of an alveolar bone which can be used for placing an implant in a bone height planting plan is the distance from the apex of the alveolar ridge to the nose bottom, the height of a bone which can be used for a maxillary anterior tooth in the dental defect area is the distance from the apex of the alveolar ridge to the maxillary sinus bottom, and the height of a bone which can be used for a mandibular anterior tooth in the dental defect area is the distance from the apex of the alveolar ridge to the lower edge of the mandible, and the height of a bone which can be used for a mandibular posterior tooth is the distance from the apex of the alveolar ridge to the chin hole or the upper edge of a mandibular nerve tube; The distance between the mesial root and the distal root of the missing tooth region refers to the distance from the mesial side position of the mesial tooth root to the mesial side position of the mesial tooth root, the distances of the positions 2,4, 6, 8, 10, 12 and 14mm below the crest of the tooth groove are measured, and the distance between the mesial root and the distal root of the missing tooth region at the narrowest part is output as a result.

Description

CBCT measurement system Technical Field The invention relates to the technical field of CBCT measurement, in particular to a CBCT measurement system. Background The key of long-term success of the planting and repairing is that the repairing is taken as a guide, the three-dimensional position and the axial direction of the implant are accurately implanted, the ideal planting and repairing effect can be obtained, the long-term success of the planting and repairing is facilitated, and various complications are easily caused by improper implantation positions of the implant. At present, the preimplantation planning can be realized through a computer-aided planning and design technology and computer-aided surgical software, and the accurate implantation of the implant is guided. Various preoperative planning requirements, such as the distance from adjacent teeth to the implant, the distance between implants, the length and diameter of the implantable implant, etc., make the virtual planning software an important tool for successful planting. At present, preimplantation planning software is required to acquire alveolar bone information and dentition information through imaging data of a patient. The CBCT image is widely applied to preoperative diagnosis of dentistry due to the advantages of lower use cost, low radiation dose, capability of performing dental jaw volume imaging and the like. Accurate reading of CBCT images is useful for determining the size (length, diameter) of the implant while avoiding damaging important anatomical structures, determining the optimal position of the implant. When an stomatist performs planning before implantation operation, the dentist can obtain implantation related parameter information such as the height, the width and the like of available bones of the alveolar bone of the tooth-missing gap and the tooth-missing area through the CBCT image. Accurate measurement of relevant parameters is critical to long-term success of oral treatment planning. However, the intelligent measurement cannot be realized in the process, manual fixed point is required by doctors, the process is complicated, time is consumed, accuracy and consistency are lacking, and factors such as individual anatomical structure differences, CBCT shooting quality and the like can influence the fixed point accuracy and the measurement analysis reliability. Disclosure of Invention The invention provides a CBCT measurement system, which performs intelligent measurement on CBCT images through a deep learning algorithm, realizes intelligent segmentation of relevant anatomical structures of a planting area and automatic measurement of relevant parameters of planting diagnosis, has reliable measurement results, and can be used for data measurement analysis of preimplantation planning. The invention provides a CBCT measurement system, which comprises a processor, a data acquisition module, an image preprocessing module, a segmentation model module, an identification module, a parameter measurement module and a result output module, wherein the processor is respectively connected with the image preprocessing module, the segmentation model module, the identification module and the parameter measurement module, the image preprocessing module is also connected with the data acquisition module, the identification module is also connected with the parameter measurement module, and the parameter measurement module is also connected with the result output module; the data acquisition module acquires CBCT image data by adopting oral cone beam CT scanning equipment; The image preprocessing module is used for preprocessing the CBCT image data acquired by the data acquisition module, wherein the preprocessing comprises resolution normalization and value standardization; the segmentation model module is used for realizing the segmentation of jawbone, maxillary sinus, nose bottom and mandibular nerve tubes by adopting nnU-Net nerve network model; the recognition module is used for recognizing and segmenting teeth by adopting an improved Mask R-CNN neural network model so as to determine tooth positions and tooth missing conditions; the parameter measurement module is used for measuring relevant parameters of planting diagnosis; and the result output module is used for outputting the results of the parameter measurement module and the identification module in an Excel table form and displaying the results at the front end. Further, in the identification module, the improved Mask R-CNN neural network model includes 4 stages, specifically: the method comprises the following steps of 1, taking a residual error network ResNet and a feature pyramid network FPN as feature extractors, and extracting features of an input image; Step2, extracting a target region ROI which possibly exists from the obtained feature map through a region generation network RPN; Inputting the ROI obtained in the step 2 into the ROI Align, and mapping the ROI into a