CN-122005140-A - Periodontal disease image-guided accurate diagnosis targeting calibration method and system

CN122005140ACN 122005140 ACN122005140 ACN 122005140ACN-122005140-A

Abstract

The application discloses a periodontal disease image guided accurate diagnosis target calibration method and a system, which relate to the technical field of medical image processing and currently propose the following scheme, wherein the method comprises the steps of obtaining three-dimensional bone density distribution data of a jaw bone area of a patient, extracting a topological coordinate set of the bone defect area and gray values of points in the topological coordinate set, obtaining a scattered signal intensity sequence of the patient along the tooth root surface and space pose data thereof, constructing an affine transformation matrix of a three-dimensional space coordinate system where the space pose data and the topological coordinate set are located based on a bimodal physical marking point preset in a patient mouth, and aligning the scattered signal intensity sequence to the topological coordinate set by utilizing the affine transformation matrix. According to the application, through spatially registering the scattered signal sequence and the three-dimensional bone density distribution data and analyzing the included angle between the gradient vector and the bone microstructure dominant vector, the limitation that the traditional CBCT depends on static bone density morphology is broken through, and the quantification of the pathological redirection of periodontal disease variable regions and the spatial association of bone microstructure dominant directions is realized.

Inventors

ZHANG ZHONGXUAN
FU XINYU
OU LINGLING
XU CHUXIN
HUANG ZHIQI
SHI WEN

Assignees

广州医科大学

Dates

Publication Date: 20260512
Application Date: 20260212

Claims (10)

1. The periodontal disease image guided accurate diagnosis target calibration method is characterized by comprising the following steps of: Acquiring three-dimensional bone density distribution data of a jawbone area of a patient, and extracting a topological coordinate set of a bone defect area and gray values of points in the topological coordinate set; Acquiring a scattered signal intensity sequence and space pose data of a patient along the tooth root surface; Based on bimodal physical marking points preset in a patient mouth, constructing an affine transformation matrix of a three-dimensional space coordinate system where space pose data and a topological coordinate set are located, and aligning a scattering signal intensity sequence to the topological coordinate set by utilizing the affine transformation matrix; Calculating a Hessian matrix for gray values of each point in the topological coordinate set, taking the ratio of the maximum characteristic value to the rest characteristic values as bone absorption activity, and taking the characteristic vector corresponding to the maximum characteristic value as a bone microstructure dominant vector; Calculating gradient vectors of the scattering signal intensity sequence after affine transformation in a three-dimensional space coordinate system, and calculating space included angles between the gradient vectors and bone microstructure dominant vectors; Integrating the bone absorption liveness, the gradient vector modular length and the space included angle into a three-dimensional feature vector, inputting the three-dimensional feature vector into a pre-trained depth self-encoder after normalizing the three-dimensional feature vector, outputting a two-dimensional latent space vector, taking the Euclidean distance between the two-dimensional latent space vector and a preset healthy tissue clustering center as an activity index, and dividing the targeting regions with different grades according to the activity index.
2. The method according to claim 1, wherein extracting the topological coordinate set and the corresponding gray values of the bone defect region comprises: calculating the mean value and standard deviation of bone densities of all voxels in the three-dimensional bone density distribution data, and subtracting the standard deviation of preset times from the mean value to serve as a primary screening threshold value; Marking all voxels with the bone density lower than a preliminary screening threshold value as preliminary screening defect voxels, and identifying and polymerizing spatially continuous preliminary screening defect voxels through a three-dimensional connected domain analysis algorithm to form a discrete candidate bone defect region; Taking the geometric center of the candidate bone defect area as an origin, and expanding a preset distance outwards along the normal vector of the surface of the candidate bone defect area to construct a spherical shell-shaped buffer area; Counting the mean value of voxel bone density in the spherical shell buffer zone, and taking the product of the mean value of voxel bone density and a preset proportion as a local judgment threshold value of the candidate bone defect area; Removing voxels with the bone density lower than the corresponding local judgment threshold value in the candidate bone defect area, and performing three-dimensional morphological closing operation and opening operation to generate a bone defect area; And extracting the space coordinates and bone density of all voxels in the bone defect region, mapping the bone density into gray values and correlating the gray values with the corresponding space coordinates to generate a topological coordinate set.
3. The method according to claim 1, characterized in that aligning the sequence of scattered signal intensities to a set of topological coordinates by means of the affine transformation matrix, in particular comprises: s1, mapping space pose data of a scattering signal intensity sequence to a three-dimensional space coordinate system where a topological coordinate set is located through an affine transformation matrix, and integrating the space pose data into the scattering signal coordinate set; S2, taking a space intersection of the scattering signal coordinate set and the topological coordinate set as an overlapping area, generating a grid point set in the overlapping area according to preset resolution, and associating a virtual scattering signal intensity value and a virtual gray value for each grid point, wherein the virtual scattering signal intensity value is generated by carrying out radial basis function interpolation on the scattering signal point set, and the virtual gray value is generated by carrying out tri-linear interpolation on the topological coordinate set; Correlating the virtual scattering signal intensity values and the virtual gray values of each grid point into data pairs, and calculating the pearson correlation coefficient based on all the data pairs; If the absolute value of the pearson correlation coefficient is lower than the preset correlation threshold, iteratively adjusting the Euler angle of the affine transformation matrix with a fixed angle step length, re-executing S1 and S2 and calculating the pearson correlation coefficient until the absolute value of the pearson correlation coefficient is not lower than the preset correlation threshold.
4. The method according to claim 1, characterized in that the calculation of the Hessian matrix for the gray values of the points in the topological coordinate set comprises in particular: traversing each topological point in the topological coordinate set, and taking the topological point as a center to extract a cube data body in a preset adjacent domain from the three-dimensional bone density distribution data; performing multi-scale filtering processing based on discrete convolution by adopting a three-dimensional Gaussian check on the cube data body, and selecting the optimal filtering scale with the largest normalized gradient norm; under the optimal filtering scale, calculating a second partial derivative of a gray value at a topological point with respect to a space coordinate of the topological point, and constructing a three-dimensional symmetric matrix by using the second partial derivative as a Hessian matrix; Three real eigenvalues of the Hessian matrix are calculated through a cyclic Jacobian iteration method, and are output in order from large to small according to absolute values.
5. The method of claim 1, wherein the pre-trained depth self-encoder comprises a first fully-connected layer, a second fully-connected layer, and a third fully-connected layer, wherein, The first full-connection layer receives the standardized three-dimensional feature vector, 128 neuron nodes of the layer respectively execute independent weight multiplication and addition operation on each dimension of the three-dimensional feature vector, apply a linear rectification function of a preset leakage coefficient and output 128-dimensional activation feature vectors; The second full-connection layer receives the 128-dimensional activation feature vector, 64 neuron nodes of the layer execute independent weight multiplication and addition operation on each dimension of the 128-dimensional activation feature vector, apply hyperbolic tangent function and output 64-dimensional activation feature vectors; and the third full-connection layer receives the 64-dimensional activation feature vector, and the 2 neuron nodes of the layer execute independent weight multiplication and addition operation on each dimension of the 64-dimensional activation feature vector to generate a 2-dimensional weighted sum vector and output the 2-dimensional weighted sum vector as a two-dimensional latent space vector.
6. The method according to claim 1, characterized in that the targeting regions of different classes are classified according to the activity index, in particular comprising: traversing the activity index of each topological point in the topological coordinate set, comparing the activity index with a preset risk level, and distributing a risk level label according to the comparison result; performing three-dimensional connected domain analysis on topological points of the labels with the same grade to generate a plurality of discrete risk areas; Calculating the projection area of each risk area, and removing the risk areas with the projection areas smaller than a preset projection threshold; Calculating the average value of the activity indexes of the internal topological points of each reserved risk area, and merging the equivalent risk areas with the spatial distance smaller than a preset interval threshold value and the difference of the average value of the activity indexes smaller than a preset difference threshold value; and taking the risk areas which remain discrete after merging as target areas, and associating the grades of the risk areas.
7. The method of claim 6, further comprising a dynamic adjustment mechanism after calculating an activity index mean of the internal topology points of each of the reserved risk zones: calculating the standard deviation of the activity indexes of each reserved risk area, and if the standard deviation is larger than a preset uniformity threshold value, removing topological points in which the absolute value of the deviation between the activity indexes and the average value in the risk area is larger than a preset deviation threshold value; Virtual outward equidistant expansion of the boundary of the current risk area to generate an expansion area, and identifying topological points which are not distributed to any risk area and have the same risk level label in the expansion area; Screening points in which the absolute value of the deviation between the activity index and the mean value of the current risk area does not exceed a preset deviation threshold value from the identified topological points, and incorporating the points into the current risk area; And updating the standard deviation of the activity index of the risk area after adjustment, and if the standard deviation of the activity index is still larger than the preset uniformity threshold value, repeating the removing and the incorporating operation until the standard deviation of the activity index meets the preset uniformity threshold value or the maximum iteration number is reached.
8. The periodontal disease image guided accurate diagnosis target calibration system, which is used for realizing the periodontal disease image guided accurate diagnosis target calibration method according to any one of claims 1 to 7, comprising: the bimodal data acquisition module is used for acquiring three-dimensional bone density distribution data of a jaw bone area of a patient, extracting a topological coordinate set of a bone defect area and gray values of points in the topological coordinate set, and acquiring a scattered signal intensity sequence of the patient along the surface of a tooth root and space pose data of the scattered signal intensity sequence; the data space alignment module is used for constructing an affine transformation matrix of a three-dimensional space coordinate system where the space pose data and the topological coordinate set are located based on bimodal physical marking points preset in the mouth of the patient, and aligning the scattering signal intensity sequence to the topological coordinate set by utilizing the affine transformation matrix; The gray value matrix construction module is used for calculating a Hessian matrix for gray values of each point in the topological coordinate set, taking the ratio of the maximum characteristic value to the rest characteristic values as bone absorption activity, and taking the characteristic vector corresponding to the maximum characteristic value as a bone microstructure dominant vector; the space parameter calculation module is used for calculating gradient vectors of the scattering signal intensity sequence after affine transformation in a three-dimensional space coordinate system and calculating space included angles between the gradient vectors and bone microstructure dominant vectors; The target area dividing module is used for integrating the bone absorption liveness, the gradient vector modular length and the space included angle into three-dimensional feature vectors, inputting the three-dimensional feature vectors into a pre-trained depth self-encoder after normalizing the three-dimensional feature vectors, outputting a two-dimensional latent space vector, taking Euclidean distance between the two-dimensional latent space vector and a preset healthy tissue clustering center as an activity index, and dividing target areas of different grades according to the activity index.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the method of any of claims 1-7 when executing the computer program.
10. A computer readable storage medium storing a computer program, which when executed by a processor performs the method according to any one of claims 1-7.

Description

Periodontal disease image-guided accurate diagnosis targeting calibration method and system Technical Field The application relates to the technical field of medical image processing, in particular to a periodontal disease image-guided accurate diagnosis targeting calibration method and system. Background In a modern periodontal disease diagnosis and treatment system, an image-guided intelligent recommendation system becomes a core support for clinical decision, a Cone Beam Computer Tomography (CBCT) is widely applied in clinic to construct a jawbone three-dimensional model, and a multi-generation image-assisted recommendation system is supported to develop, wherein a typical operation flow is as follows, after CBCT data of a jawbone face of a patient is acquired, a bone density distribution map is generated through gray threshold segmentation and three-dimensional reconstruction; and finally matching the score with a preset rule base, and outputting a recommendation list comprising coordinates, intervention degree and priority ordering of a targeted intervention area. Along with the upgrade of the accurate medical requirements, the bottom image analysis model still samples the static morphological characterization, and the time sequence biological progress and microstructure space guiding evolution of bone resorption cannot be captured, so that the accuracy and the severity of the target area suffer from bottlenecks. In periodontal disease image analysis, the prior art has the following limitations that the traditional imaging method (such as CBCT) can only statically present the bone defect form, so that the dynamic activity of the bone absorption process and the main direction of the bone microstructure are difficult to quantify, active lesions and old lesions cannot be effectively distinguished, and the definition of a target area depends on the empirical judgment of doctors on the bone defect form to influence the definition of a subsequent intervention area and the setting of the intervention degree, so that the periodontal disease image-guided accurate diagnosis target calibration method and system are provided to solve the problem. Disclosure of Invention In order to solve the technical problems, the technical scheme solves the problems in the background technology by providing a periodontal disease image-guided accurate diagnosis targeting calibration method and system. In order to achieve the above object, the technical scheme of the present invention is as follows: In a first aspect, the present application provides a periodontal disease image guided accurate diagnostic targeting calibration method comprising: Acquiring three-dimensional bone density distribution data of a jawbone area of a patient, and extracting a topological coordinate set of a bone defect area and gray values of points in the topological coordinate set; Acquiring a scattered signal intensity sequence and space pose data of a patient along the tooth root surface; Based on bimodal physical marking points preset in a patient mouth, constructing an affine transformation matrix of a three-dimensional space coordinate system where space pose data and a topological coordinate set are located, and aligning a scattering signal intensity sequence to the topological coordinate set by utilizing the affine transformation matrix; Calculating a Hessian matrix for gray values of each point in the topological coordinate set, taking the ratio of the maximum characteristic value to the rest characteristic values as bone absorption activity, and taking the characteristic vector corresponding to the maximum characteristic value as a bone microstructure dominant vector; Calculating gradient vectors of the scattering signal intensity sequence after affine transformation in a three-dimensional space coordinate system, and calculating space included angles between the gradient vectors and bone microstructure dominant vectors; Integrating the bone absorption liveness, the gradient vector modular length and the space included angle into a three-dimensional feature vector, inputting the three-dimensional feature vector into a pre-trained depth self-encoder after normalizing the three-dimensional feature vector, outputting a two-dimensional latent space vector, taking the Euclidean distance between the two-dimensional latent space vector and a preset healthy tissue clustering center as an activity index, and dividing the targeting regions with different grades according to the activity index. In a second aspect, the present application provides a periodontal disease image guided accurate diagnostic target calibration system for implementing the periodontal disease image guided accurate diagnostic target calibration method described in any one of the above, comprising: the bimodal data acquisition module is used for acquiring three-dimensional bone density distribution data of a jaw bone area of a patient, extracting a topological coordinate set of a bone