Method for quantitative evaluation of bone augmentation before and after periodontitis treatment based on curved tomographic slices

Abstract

The application is a kind of periodontitis treatment before and after bone increment quantitative evaluation method based on curved surface tomography, which firstly extracts the tooth feature point pair of the evaluated part in the oral curved surface tomography obtained twice before and after treatment, automatically extracts the feature point of the edge point, matches the feature point of the edge point in the two images; a registration model is constructed to transform the post-treatment oral curved surface tomography; according to the position of the feature point pair, the post-treatment oral curved surface tomography after registration is locally sliced to obtain the registered slice, and the image gray of the registered slice is segmented to obtain a plurality of consistent connected regions, the alveolar bone change target position is automatically selected, and the alveolar bone increment area is calculated by using the number of region pixels and the real resolution of the curved surface tomography. The application can accurately transform the post-treatment picture to the locally aligned position with the pre-treatment picture, automatically extract the change area, and then obtain the accurate alveolar bone increment area.

Description

TECHNICAL FIELD

[0001] The application relates to a periodontitis pre-and post-treatment bone increment quantitative evaluation method based on a curved surface tomogram, and belongs to the novel imaging technical field of alveolar bone increment evaluation. BACKGROUND

[0002] The change of alveolar bone height is an important clinical index for influencing the periodontitis staging and grading, and is also one of important clinical evaluation indexes for evaluating the treatment effect. The loss of alveolar bone can cause tooth sector displacement in the anterior tooth area, and affects the appearance and cutting occlusion function; the tooth loss caused by alveolar bone absorption in the posterior tooth area seriously affects the mastication function and efficiency. The certain degree of bone absorption causes the formation of deep periodontal pocket, and the bacteria and toxins in the pocket cause the continuous development of periodontitis, the increase of inflammation level, and the further absorption of alveolar bone, forming a vicious cycle. It has been shown by the present research that periodontal treatment can improve the periodontal inflammation level, promote the repair and regeneration of alveolar bone to different degrees.

[0003] Imaging evaluation is an important part of periodontal clinical evaluation, and can provide information on the past destruction of periodontal hard tissue, and has important value for longitudinal measurement of progressive bone loss and alveolar bone regeneration. In the diagnosis and treatment process of periodontitis, in order to more intuitively observe the absorption and regeneration of alveolar bone, oral curved surface tomography is often needed to compare and observe the reconstruction process of alveolar bone in multiple parts of the whole mouth. The oral curved surface tomogram not only gives the overall cognition of the alveolar bone condition of the whole mouth of the doctor and patient, but also has important significance for judging the alveolar bone reconstruction before and after periodontal treatment, the probability of tooth preservation, prognosis evaluation, and the formulation of the later periodontal treatment scheme. In addition, the oral curved surface tomogram is simple to shoot, and the price is easy to accept, and has good accuracy and repeatability. These advantages are one of important factors for its wide application. In addition, compared with CBCT and other 3D images, the oral curved surface tomogram has the advantages of simple imaging and low cost, and it is easier to realize the change detection of alveolar bone structure as an important auxiliary means for periodontitis monitoring, therefore, the change detection research of the oral curved surface tomogram has important significance.

[0004] Registration is a prerequisite for multi-temporal image sequence processing. The concept of registration is defined as the image processing method of referencing the coordinate system of different images to one of the standard images or existing images, and performing coordinate transformation to make the objective positions of each pixel in the image as consistent as possible. In the field of medical image processing, according to the attributes of the input image, image registration can be divided into single / multi-modal registration, single patient registration, multi-patient registration, etc.; according to the dimension of the input image, it can be divided into 3D-3D registration, 2D-2D registration and 2D-3D registration, etc.; according to the image registration transformation model, it can be divided into rigid model registration, affine transformation model registration, deformable model registration, etc.; image registration has been applied in many medical research fields, such as registering brain MRI / CT images with standard brain models to obtain the partition of each region of the actual image, thereby assisting the diagnosis and efficacy observation of brain diseases; registering pre-treatment and post-treatment CT lung images to track and observe lung tissue during the breathing process, thereby assisting the treatment of lung diseases. However, at present, there are few registration methods specially designed for oral images, and it is often difficult to directly apply general image registration methods to oral curved tomography due to the inclusion of the movable mandible and the central rotation imaging mode.

[0005] After the images are completely registered, an image segmentation method is needed to accurately extract the region where the alveolar bone has changed. The extraction of the image region of interest has always been an important research direction in image processing. The current image region of interest extraction method generally uses image segmentation means, and the image segmentation method can be divided into two categories: traditional segmentation method and segmentation method based on deep learning. The traditional segmentation method generally uses texture features or gray level features to segment the image using the uniformity and homogeneity of the same region, which has the advantages of being convenient and fast, and does not need or only needs a few labeled images as prior knowledge. However, the extraction effect depends on the effectiveness and accuracy of the manually designed features, and the general applicability is often low; the segmentation method based on deep learning uses deep neural networks to automatically extract and learn features, which can break free from the constraints of human-designed features, but the premise is that more labeled samples are needed as training samples, and the current neural network volume is relatively large, the training network requires a long time, and the labeling cost is high. At present, there are still few methods for extracting the alveolar bone change region of the oral curved tomography. Since the change of the alveolar bone often only exists in the region with severe alveolar bone recession, this region often has high discontinuity, so the global region of interest extraction method is generally difficult to directly apply. SUMMARY

[0006] To solve the problems in the prior art, the present application provides a method that can transform the post-treatment curved tomography to a position that is locally accurately aligned with the pre-treatment curved tomography, and then obtain accurate alveolar bone augmentation quantification.

[0007] In order to achieve the above object, the technical scheme of the present application is as follows: a periodontitis pre-and post-treatment bone increment quantitative evaluation method based on curved surface tomography, comprising the following steps:

[0008] Step one, pretreatment of the pre-and post-treatment oral curved surface tomography;

[0009] Step two, extraction of M tooth feature point pairs in the alveolar bone change evaluation area of the oral curved surface tomography as the basic feature point pairs, and extraction of the edge point set N is the total number of edge points;

[0010] Step three, extraction of the edge point set feature points of each edge point, and matching of the feature points of the edge points of the pre-and post-treatment curved surface tomography;

[0011] Step four, merging of the obtained edge point feature point matching results and the basic feature point pairs to generate a coarse matching point pair set;

[0012] An initial registration model is established as follows:

[0013]

[0014] Where (x r ,y r ) is the pixel point coordinate in the pre-treatment tomography image, and (x,y) is the pixel point coordinate in the post-treatment tomography image;

[0015] Each time, a number of feature point pairs are selected from the coarse matching point pair set at random, and a transformation matrix H is calculated; then the number of matching points satisfying the following formula is calculated:

[0016]

[0017] Where t is a self-determined threshold;

[0018] The transformation matrix H corresponding to the largest number of matching points satisfying the above formula is used to substitute into the initial registration model to obtain a final registration model;

[0019] Step five, using the final registration model to transform the post-treatment oral curved surface tomography to obtain a registered post-treatment oral curved surface tomography aligned with the pre-treatment oral curved surface tomography;

[0020] Step six, local slicing of the registered post-treatment oral curved surface tomography according to the position of the basic feature point pairs to obtain a registered slice, and image gray scale segmentation of the registered slice;

[0021] Step seven, after image gray scale segmentation, a plurality of consistency connected regions are obtained, a target region of alveolar bone change is selected, and a pixel number p of the alveolar bone change region is obtained;

[0022] The alveolar bone increment area S is calculated by using the pixel number p and the real resolution R of the curve tomography slice:

[0023] S = p x R.

[0024] The further design of the above technical solution is that: the pre-processing in step one includes histogram standardization and median filter processing. The selected feature point pairs in step two include the tooth tip of the alveolar bone change evaluation region, the high point of the crown shape of the mesial-distal surface, the enamel-dental cementum junction, the root tip and the root bifurcation region and other easily recognizable dental anatomical shapes.

[0025] The step three uses the scale invariant feature transform descriptor to extract the edge point set The feature points of each edge point are extracted, and the two-near-neighbor method is used to match the feature points of the edge points of the two curve tomography slices before and after treatment under the following constraints:

[0026] 1) The ratio of the Euclidean distance of the matched edge point to the nearest neighbor point to the Euclidean distance to the second nearest neighbor point is less than a threshold value λ;

[0027] 2) The distance of the feature vector of the edge point to all basic feature points is calculated, and the nearest neighbor basic feature points corresponding to the mutually matched edge point pairs are matched with each other;

[0028] 3) The Euclidean distance between the matched edge point and the nearest basic feature point should be less than a threshold value l.

[0029] The generalized Ostu method is used for image gray scale segmentation in step six, and the specific method is as follows:

[0030]

[0031] Wherein, k1, k2, …, k N N segmentation thresholds, N i is the pixel number contained in the i-th section under the current segmentation, N is the total number of pixels, μ i is the mean value of all pixel gray values of the i-th section under the current segmentation, μ is the gray mean value of all pixels.

[0032] The beneficial effects of the present application are:

[0033] The registration part of the oral curve tomography slice in the method can make the local registration precision error within 1 pixel, so as to ensure that the post-treatment picture is transformed to a position locally accurately aligned with the pre-treatment picture. The subsequent image segmentation can extract a relatively accurate alveolar bone change region and give the corresponding specific area.

[0034] The method of the present application is based on a conventional image automatic segmentation method, thus no pre-training of the model is required, and the segmentation and registration can give results within 1s, and the image processing results can be output in real time according to the requirements. The quantitative evaluation results obtained can more accurately and intuitively compare and evaluate the loss and increase of alveolar bone volume, and provide a diagnosis and treatment basis for long-term treatment plan and prognosis of periodontal treatment. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 is a flowchart of the method of the present application;

[0036] Figure 2 is a pre-treatment oral curved surface tomogram of a patient;

[0037] Figure 3 is a post-treatment oral curved surface tomogram of a patient;

[0038] Figure 4 is a superimposed result of the pre-treatment and post-treatment oral curved surface tomograms of a patient;

[0039] Figure 5 is a superimposed result of the pre-treatment and post-treatment oral curved surface tomograms of a patient;

[0040] Figure 6 is an alveolar bone increment detection result between the second premolar and the first molar in the posterior region of a patient. DETAILED DESCRIPTION

[0041] The present application will be described in detail below in conjunction with the drawings and specific embodiments.

[0042] EMBODIMENT

[0043] The alveolar bone increment quantitative method based on oral curved surface tomograms before and after periodontal treatment of the present embodiment is specifically as shown in Figure 1

[0044] Taking two oral curved surface tomograms of a patient before and after periodontal treatment with a three-year interval as an example, the two images are as shown in Figure 2 and Figure 3 According to the judgment of a doctor, there is obvious bone mass growth between the second premolar and the first molar in the posterior region, which indicates the effectiveness of the treatment. However, the result after superimposing the two images is as shown in Figure 4 It can be seen that the above two pictures cannot be directly aligned tooth by tooth, and it is difficult to quantitatively evaluate the increment of alveolar bone. Taking the evaluation of the alveolar bone change between the second premolar and the first molar in the posterior region as an example, the specific scheme of the present method is as follows:

[0045] Step one, pre-treat the two curved surface tomograms to improve the image visual effect and align the gray scale distribution, the pre-treatment includes histogram standardization and median filter processing.​

[0046] Step two, manually extract 8 tooth feature point pairs of the desired alveolar bone change evaluation area, including the cement-enamel junction of two teeth, the root tip and the first molar root bifurcation area. In addition, use edge extraction method to extract edge point set N is the total number of edge points.

[0047] Step three, take the manually extracted feature point pairs as the basis feature point pairs, and match the point pairs with constraints to improve registration accuracy. Use the scale-invariant feature transform (SIFT) descriptor to extract the feature points of each edge point, and use the two-neighbor method for matching, but at the same time, the following constraints are made:

[0048] 1) The ratio of the Euclidean distance between the matched edge point and the nearest neighbor point to the Euclidean distance between the point and the second nearest neighbor point is less than the threshold value λ = 0.4;

[0049] 2) Calculate the distance between the feature vector of the edge point and all basis feature points. The nearest neighbor basis feature point corresponding to the matching point should match with the nearest neighbor basis feature point corresponding to the matching point;

[0050] 3) The Euclidean distance between the matched edge point and the nearest manually extracted feature point in the graph should be less than the threshold value l = 20.

[0051] Step four, merge the obtained edge point matching results with the basis feature point pairs to generate a rough matching point pair set. Use the random sample consensus (RANSAC) method for further screening of matching point pairs and calculation of registration model. Since the oral curved surface tomogram is taken for the mandibular bone structure, which can be locally regarded as a rigid perspective transformation, the initial registration model is set as a perspective model:

[0052]

[0053] Where (x r ,y r ) is the pixel point coordinate in the pre-treatment image, and (x, y) is the pixel point coordinate in the post-treatment image.

[0054] The specific process of the RANSAC method is as follows: randomly select 4 groups of feature points from the rough matching point pair set each time, use the singular value decomposition (SVD) method to calculate the transformation matrix H, and calculate the number of matching points that satisfy the following formula:

[0055]

[0056] where t is a self-defined threshold, in this example, t = 1. The number of feature point pairs is equal to the number of times of calculation. After multiple calculations, the transformation matrix H corresponding to the maximum number of points satisfying the above formula is selected, and the final registration model is obtained by substituting the transformation matrix H into the initial registration model.

[0057] Step five, using the final registration model to transform the post-treatment oral curve surface film to obtain the registered post-treatment oral curve surface film completely aligned with the pre-treatment image, as shown in FIG. 5. It can be seen that the treatment site near the posterior teeth area (in the box) achieves accurate registration. After the above operation, the post-treatment picture is registered with the pre-treatment picture in this embodiment, and the local registration accuracy error is within 1 pixel, ensuring that the post-treatment picture is transformed to a position accurately registered with the pre-treatment picture. Figure 5

[0058] Step six, according to the location of the basic feature point pairs, the local slice processing of the curve surface film is performed, and the generalized Ostu method is used to segment the image gray scale of the registered slice:

[0059]

[0060] where k1, k2, …, k N N are N segmentation thresholds, and in this example, N = 5. N i is the number of pixels contained in the i-th segment under the current segmentation, N is the total number of pixels, μ i is the mean value of the gray scale of all pixels in the i-th segment under the current segmentation, and μ is the gray scale mean value of all pixels.

[0061] Step seven, after image segmentation, a plurality of consistent connected regions are obtained, and at this time, the user interactively selects the target position of the alveolar bone change, that is, the alveolar bone change region information of interest, including the region contour and the region pixel number, is obtained, so as to mark in the original image, and the alveolar bone change contour obtained is as shown in FIG. 6. It can be seen that this embodiment can accurately detect the contour of the alveolar bone change position, so as to measure and obtain the area information. Figure 6

[0062] The alveolar bone increment area S is calculated by using the region pixel number p and the real resolution (a ruler can be added during shooting to calculate) R (cm 2 / pixel) of the curve surface film:

[0063] S = p × R

[0064] In this example, S = p × R = 709 × 1.33 × 10 -4 = 0.0942 cm 2 .

[0065] ​​The embodiment can extract a relatively accurate alveolar bone change region and give a corresponding specific area, which basically conforms to the actual alveolar bone increment, and the error is very small.

[0066] The technical solutions of the present application are not limited to the above embodiments, and any technical solution obtained by equivalent replacement falls within the scope of the present application.

Claims

1. A method for quantitatively assessing bone gain before and after periodontitis treatment based on curved surface tomographic images, characterized in that: Includes the following steps: Step 1: Preprocess the panoramic radiographs of the oral cavity before and after treatment; Step 2: Extract M pairs of tooth feature points from the alveolar bone change assessment area in the panoramic radiograph of the oral cavity as the basic feature point pairs, and extract the edge point set. N is the total number of edge points; Step 3: Extract the set of edge points Feature points of each edge point in the middle, and match feature points of edge points of two curved tomographic slices before and after treatment; Step 4: Merge the feature point matching results of the obtained edge points with the basic feature point pairs to generate a coarse matching point pair set; The following initial registration model is established: Among them, (x r ,y r (x, y) represents the pixel coordinates in the tomographic image before treatment, and (x, y) represents the pixel coordinates in the tomographic image after treatment. Each time, several sets of feature point pairs are randomly selected from the coarse matching point pair set, and the transformation matrix H is calculated; then, the number of all matching points that satisfy the following formula is calculated: Where t is a user-defined threshold; The transformation matrix H corresponding to the maximum number of matching points that satisfies the above formula is substituted into the initial registration model to obtain the final registration model. Step 5: Use the final registration model to transform the post-treatment oral panoramic tomographic image to obtain a registered post-treatment oral panoramic tomographic image that is aligned with the pre-treatment oral panoramic tomographic image. Step 6: Based on the location of the basic feature points, perform local slicing on the registered post-treatment oral surface tomographic film to obtain the registered slices, and then perform image grayscale segmentation on the registered slices. Step 7: After image grayscale segmentation, multiple connected regions with consistency are obtained. The target region of alveolar bone change is selected, and the number of pixels p in the alveolar bone change region is obtained. The incremental alveolar bone area S is calculated using the number of pixels p in the region and the true resolution R of the tomographic slice: S = p × R.

2. The method for quantitatively assessing bone gain before and after periodontitis treatment based on curved tomographic images according to claim 1, characterized in that: The preprocessing in step one includes histogram standardization and median filtering.

3. The method for quantitatively assessing bone gain before and after periodontitis treatment based on curved tomographic images according to claim 1, characterized in that: The feature points selected in step two include the cusps in the alveolar bone change assessment area, the highest point of the mesial and distal crown shape, the cementoenamel junction, and the root apex and root bifurcation area.

4. The method for quantitatively assessing bone gain before and after periodontitis treatment based on curved tomographic images according to claim 1, characterized in that: In step three, the scale-invariant feature transform descriptor is used to extract the set of edge points. The feature points of each edge point in the tomographic images are identified, and the two nearest neighbors method is used to match the feature points of the edge points of the two curved tomographic images before and after treatment under the following constraints; 1) The ratio of the Euclidean distance from the matched edge point to the nearest neighbor to the Euclidean distance to the second nearest neighbor is less than the threshold λ; 2) Calculate the distance between the descriptive feature vector of the edge point and all basic feature points. The nearest neighbor basic feature points of mutually matching edge points are mutually matched. 3) The Euclidean distance between the matched edge point and the nearest basic feature point should be less than the threshold l.

5. The method for quantitatively assessing bone gain before and after periodontitis treatment based on curved tomographic images according to claim 1, characterized in that: Step six uses the generalized Ostu method to segment the image grayscale, as detailed below: Where k1,k2,…,k N For N segmentation thresholds, N i Let μ be the number of pixels in the i-th segment under the current segmentation, N be the total number of pixels, and μ be the number of pixels in the segment. i is the average grayscale value of all pixels in the i-th segment under the current segmentation, and μ is the average grayscale value of all pixels.

Images