KR101381061B1 - A method for extracting the structure of teeth automatically from computed tomography image - Google Patents

Abstract

The present invention helps precise orthodontic treatment by increasing the matching rate between the tooth information obtained by automatically extracting the dental structure of the subject through the image processing from the CT image of the subject used in the orthodontic treatment and the actual tooth of the subject, orthodontic treatment In this paper, we provide an automatic extraction method of dental structures from CT images that can quantitatively analyze tooth movement and tooth changes.

Description

A method for extracting the structure of teeth automatically from computed tomography image}

The present invention relates to a method for automatically extracting dental structures from CT images. It is about.

Currently, the orthodontic diagnosis and treatment system has been produced by intra-oral model to determine the shape, size, degree of crowning, occlusal relationship between the upper and lower teeth.

That is, since it depends on the accumulated experience in predicting and performing the ideal occlusion from the patient's information, there is no way to quantify the experience in a systematic manner depending on the operator.

More specifically, orthodontic force is performed by using various types of brackets, various materials, archwires of various thicknesses, and various mini implants, orthodontic force, but the current diagnosis and treatment plan (orthodontic) There was no way to measure quantitative correction in the treatment plannig protocol.

However, there is a method of firstly aligning the anterior teeth on a certain line with excessive orthodontic force and then gathering them into normal occlusion, but this does not mean that all teeth move along the optimal path.

Therefore, unnecessary tooth movement occurs until reaching the normal occlusion, which causes a long treatment period and at the same time increases the possibility of side effects such as root resorption.

In addition, since the treatment and correction is made over a long time there was a problem that requires a lot of patience and pain for both the clinician and the patient.

An object of the present invention for solving the above problems is to provide an automatic extraction method of the dental structure from the CT image to automatically extract the dental structure of the subject through image processing in the CT image of the subject used in the orthodontic treatment.

It is also an object of the present invention to obtain dental information and subjects obtained through infill processing, anisotropic diffusion filter processing, binarization processing, erosion processing, automated selection processing, expansion processing, masking processing and morphological filtering performed in an image processing apparatus. It is to provide an automatic extraction method of dental structure from CT images to help precise orthodontic treatment by increasing the matching rate with actual teeth.

In order to solve the above problems, the automatic extraction method of the dental structure from the CT image according to the present invention comprises the steps of: (a) obtaining a CT image of the subject through a CT imaging device; (b) an image processing step of selecting dental information through correction and noise removal of the obtained CT image transmitted to the image processing apparatus; (c) extracting a dental image from the imaged CT image; And (d) outputting the dental structure of the subject from the image output apparatus through a second polynomial fitting algorithm based on the extracted tooth image.

Preferably, in the step (b), (b1) the degree of filtering in each direction through an infilled CT image for deleting and correcting only a portion corresponding to the pulp of the tooth in the selected CT image among the obtained CT images Characterized in that it comprises a; differently determined by processing with an anisotropic diffusion filter to remove noise.

Preferably, the step (b), after the step (b1), (b2) binarizing the CT image treated with the anisotropic diffusion filter; (b3) eroding the binarized CT image; And (b4) selecting only dental information by performing an automatic selection process on the eroded CT image.

Preferably, the step (c) comprises the steps of: (c1) dilatating the automated selection processed dental image after the step (b); (c2) masking the expanded dental image; (c3) refilling the masked tooth image; And (c4) performing morphological filtering on the infill-treated dental image.

Preferably, the step (d) comprises: (d1) extracting edge information of the morphologically filtered dental image after the step (c); (d2) extracting a dental structure by the second order polynomial fitting algorithm based on the edge information of the dental image; And (d3) outputting the extracted dental structure through the image output device.

(단, I(i_f, t)는 반복스텝 t에서의 영상이미지, dvi는 발산연산자,

는 국부적인 영상 그라디언트, C(i_f, t)는 영상 그라디언트 크기의 단조감소함수(monotonically decreasing function)의 확산함수)의 열전도 방정식인 비등방성 프로세스를 연산하여 상기 치아영상의 노이즈를 제거하고, 그리고 상기 시간(t)과 상기 치아영상의 노이즈는 비례관계에 있는 것을 특징으로 한다.Preferably, the anisotropic diffusion filter treatment of the step (b1),

Where I (i _f , t) is an image image at repeat step t, dvi is an divergence operator,

Is a local image gradient, C (i _f , t) is an anisotropic process that is a thermal conductivity equation of monotonically decreasing function of the image gradient size, and removes noise of the dental image, and The time t and the noise of the dental image are in a proportional relationship.

의 확산 프로세스를 연산하여 획득되고, 그리고 상기 확산 프로세스는 최근접한 4개의 이웃영상에 의한 흐름이 같은 양에 의한 영상(i_f)에서 각각의 픽셀(x, y)의 업데이트로 이루어지는 것을 특징으로 한다.Preferably, all the images for selecting the diffusion function,

The diffusion process is obtained by calculating a diffusion process, and the diffusion process is characterized in that the flow by the four nearest neighboring images consists of updating each pixel (x, y) in the image i _f by the same amount. .

(단,

는 반복상수,

는 최근접한 이웃영상의 차이를 이용하여 계산된 국부적인 그라디언트의 방향, 최근접한 이웃영상의 차이는

인 반복스텝 t에서 동쪽방향으로

, 서쪽방향으로

, 남쪽방향으로

, 북쪽방향으로

를 의미함)의 식을 연산하여 획득하는 것을 특징으로 한다.Preferably, the four nearest neighbor images,

(only,

Is the iteration constant,

The direction of the local gradient calculated using the difference between the nearest neighbor images, and the difference between the nearest neighbor images

In the repeat step t east

, West

, Southward

, North

It is characterized by obtaining by calculating the expression of).

Preferably, the step (a) comprises the steps of: (a1) acquiring a plurality of dental images in a CT imaging apparatus; And (a2) selecting one of the plurality of dental images.

The present invention as described above has the effect of providing the basic data for the prediction of the dental structure by automatically extracting the dental structure of the subject through image processing from the CT image of the subject used for orthodontic treatment.

In addition, the present invention is the dental information obtained by the infill treatment, anisotropic diffusion filter processing, binarization processing, erosion treatment, automated selection processing, expansion processing, masking processing and morphological filtering performed in the image processing apparatus By increasing the matching rate with the teeth it is effective to help precise orthodontic treatment.

In addition, the present invention has an effect capable of quantitatively interpreting tooth movement and the amount of change of teeth in orthodontic treatment as described above.

1 is a flowchart showing an automatic extraction method of a dental structure from a CT image according to an embodiment of the present invention;
2 is a slice photograph showing a tooth image obtained by a CT image processing apparatus of an automatic extraction method of a dental structure from a CT image according to an embodiment of the present invention;
3 is a photograph showing a selected tooth image of the plurality of tooth images of FIG.
4 is a photograph showing an infill-treated CT image of FIG.
5 is a photograph showing a CT image of the anisotropic diffusion filter of the infill-treated CT image of FIG.
FIG. 6 is a photograph showing a binarized CT image of the CT image of the anisotropic diffusion filter of FIG. 5;
Figure 7 is a photograph showing the erosion treatment CT image of the binarized CT image of Figure 6,
Figure 8 is a photograph showing the tooth image of the automated selection of the erosion treatment CT image of Figure 7,
FIG. 9 is a photograph showing a tooth image in which the automated selection processed tooth image of FIG. 8 is expanded;
10 is a photograph showing a dental image masked treatment of the expansion dental image,
11 is a photograph showing a dental image masked tooth image is infilled again,
12 is a photograph showing a tooth image in which the infill-treated tooth image is morphologically filtered,
FIG. 13 is a photograph of a morphologically filtered tooth image automatically extracting only edge information of a tooth by an Otsu algorithm, and
14 is a graph of tooth structure extracted from edge tooth information by a second order polynomial fitting algorithm.

The constituent steps constituting the automatic extraction method of the dental structure from the CT image of the present invention may be used integrally or separately separated as necessary. In addition, some configuration steps may be omitted depending on the use form.

A preferred embodiment of the automatic extraction method of the dental structure from the CT image according to the present invention will be described with reference to FIGS. In this process, the thicknesses of the lines and the sizes of the components shown in the drawings may be exaggerated for clarity and convenience of explanation. In addition, the terms described below are defined in consideration of the functions of the present invention, which may vary depending on the intention or custom of the user, the operator. Therefore, the definitions of these terms should be described based on the contents throughout this specification.

Hereinafter, an automatic extraction method of a dental structure from a CT image according to an embodiment of the present invention will be described with reference to FIGS. 1 to 14.

According to an embodiment of the present invention, a method for automatically extracting a dental structure from a CT image includes (a) acquiring a CT image of a subject through a CT imaging apparatus (S10 to S20), and (b) acquiring the image transmitted to an image processing apparatus. Image processing step of correcting the CT image, the noise removal of the corrected CT image, the binarization processing of the noise-removed CT image, the erosion treatment of the binarized CT image and the automatic selection processing of the eroded CT image (S30 ~ S70) ), (c) extracting a dental image from the image-processed CT image (S80 ~ S120), and (d) an image output of the dental structure of the subject through a second polynomial fitting algorithm based on the extracted tooth image Outputting from the device (S130 ~ S140).

First, the step (a) may include (a1) acquiring a plurality of dental images in the CT imaging apparatus 100 (S10), and (a2) selecting a single dental image in the image processing apparatus 200 (S20). ).

In the step (a1), the CT image of the subject used for calibration includes about 460 slices as shown in FIG. 2, and the head image of the subject may be three-dimensionally modeled by the program (Vworks). have.

In this case, since about 460 slices include not only images including teeth but also images, the process of selecting an image including teeth should be performed first.

Next, in the step (a2), the slice including the tooth information starting from the head of the subject corresponds to Nos. 290 to 380, and among them, slices 322/323 or 348, which are slices containing the maxillary or mandibular dental structures. Select slice / 349 (see FIG. 3).

Next, the step (b), (b1) the degree of filtering in each direction through the CT image subjected to infill (S30) to delete only the portion corresponding to the pulp of the tooth in the selected CT image of the obtained CT image The method may further include the step of processing an anisotropic diffusion filter to remove noise by differently determining (S40).

That is, in the step (b1), it would be ideal if only the enamel or dentin portion of the tooth is seen in the selected image as shown in FIG. 4, but the teeth are different from each other in pulp and pulp because the teeth are not uniform in the actual human teeth. The same part comes out black and looks as if there are holes in the teeth.

Therefore, since these parts act as noise in the image processing, the efficiency of tooth extraction is lowered. Thus, such a part is deleted using an infill function.

Next, the step (b), after the step (b1), (b2) binarizing the CT image processed by the anisotropic diffusion filter (S50), (b3) erodes the binarized CT image Processing (S60) and (b4) further comprises a step (S70) of selecting only the dental information by the automated selection process of the eroded CT image.

The anisotropic diffusion filter in step (b2) compensates for the problem that the image is blurred in the process of smoothing the boundary of the image in most conventional noise removing filters.

의 열전도 방정식을 연산하여 치아영상의 노이즈를 제거하게 되는데, 보다 구체적인 설명은 다음과 같다.Anisotropic diffusion filter processing of the step (b2) to solve the above conventional problems,

The noise of the tooth image is removed by calculating the heat conduction equation of. More specific description is as follows.

The basic principle is that if an isotropic filter performs filtering that is equally smooth in all directions, the anisotropic filter analyzes the information in the image and determines the degree of filtering in each direction differently.

This anisotropic diffusion filter was originally derived from the conduction of heat and when a hot object is placed in the room, the heat spreads out over time, and the thermal conductivity equation transfers heat from the heat source to the surroundings over time. Is an equation describing the process.

In the CT image, the noise of the image is treated as a heat source, and the thermal conductivity equation makes the noise thinner, resulting in the effect of removing the noise.

However, since the thermal conductivity equation describes isotropic diffusion without direction, blurry blurring occurs.

Therefore, the Laplace term of the image, which diffuses into the space for preserving the boundary, is combined with the eigenvectors of the gradient covariance matrix, the eigenvalues, and the hessian matrix of the image, to form a constant shape. The thermal conductivity equation is as follows.

는 국부적인 영상 그라디언트, C(i_f, t)는 영상 그라디언트 크기의 단조감소함수(monotonically decreasing function)의 확산함수, 확산함수의 평가(evaluate)에 의한 경계는 선택적으로 부드러워지거나 선명해진다.Where I (i _f , t) is an image image at repetition step t, dvi is a divergence operator,

Is the local image gradient, C (i _f , t) is the monotonically decreasing function of the image gradient size, and the boundary by evaluating the diffusion function is selectively softened or sharpened.

의 유사방정식을 이용하여 연산하고, 그리고 시간(t)와 치아영상의 노이즈는 비례관계에 있다.By selecting the gradient function of the image that preserves the boundary, it was proposed by Perona and Malik as follows, and the transfer coefficient of the thermal conductivity equation is

Calculation is performed using the similarity equation of, and the time t and the noise of the dental image are proportional to each other.

In addition, all images are obtained by Equation 2 below to select a diffusion function.

K is a diffusion constant or a flow constant. The diffusion process consists of updating each pixel (x, y) in the image i _f by the same amount of flow by the four nearest neighbors. Recently, four neighboring images (neighbors) can be expressed as shown in [Equation 3].

는 반복상수이고

는 최근접한 이웃영상의 차이를 이용하여 계산된 국부적인 그라디언트의 방향이다. 최근접한 이웃영상의 차이는

인 반복스텝 t에서 동쪽방향으로

, 서쪽방향으로

, 남쪽방향으로

, 북쪽방향으로

를 말한다.but,

Is an iteration constant

Is the direction of the local gradient calculated using the difference between the nearest neighbor images. The difference between the nearest neighbors is

In the repeat step t east

, West

, Southward

, North

.

As a result, the performance of the anisotropic diffusion filter is determined by the repetition step t and the diffusion constant K. When the repetition step is increased, the image is blurred, and when the diffusion constant K is larger, the image is also blurred.

At this time, K is a fixed constant and if the value is low, the noise is hardly eliminated. On the contrary, if the value is too high, the image looks abnormal.

That is, t, which stands for time, describes the time at which the noise is spread (the number of repetitions of the operation). As t increases, the noise is removed more and more.

The CT images of the examinee subjected to the anisotropic diffusion filter treatment calculated by Equations 1 to 3 are shown in FIG. 5.

On the other hand, the diffusion constant is determined by the relationship between the degree of diffusion and the degree of local boundary, and in this patent, t = 10 and K = 30 were selected after many trial and error.

In addition, the above process implemented [Equation 2] and [Equation 3] in MATLAB, the m-file (m-file) for this is as follows.

% 2D Convolution Masking

hN = [0 1 0; 0-1 0; 0 0 0;

hS = [0 0 0; 0-1 0; 0 1 0];

hE = [0 0 0; 0-1 1; 0 0 0;

hNE = [0 0 1; 0-1 0; 0 0 0;

hSE = [0 0 0; 0-1 0; 0 0 1];

hSW = [0 0 0; 0-1 0; 1 0 0];

hNW = [1 0 0; 0-1 0; 0 0 0;

Distance information at% center pixel

dx = 1;

dy = 1;

dd = sqrt (2);

Calculation of% anisotropic diffusion filter

% im video

for t = 1:10

% Finite differences. [imfilter (.,., 'conv') can be replaced by conv2 (.,., 'same')]

nablaN = imfilter (im, hN, 'conv');

nablaS = imfilter (im, hS, 'conv');

nablaW = imfilter (im, hW, 'conv');

nablaE = imfilter (im, hE, 'conv');

nablaNE = imfilter (im, hNE, 'conv');

nablaSE = imfilter (im, hSE, 'conv');

nablaSW = imfilter (im, hSW, 'conv');

nablaNW = imfilter (im, hNW, 'conv');

cN = exp (-(nablaN / Kappa). ^ 2);

cS = exp (-(nablaS / Kappa). ^ 2);

cW = exp (-(nablaW / Kappa). ^ 2);

cE = exp (-(nablaE / Kappa). ^ 2);

cNE = exp (-(nablaNE / Kappa). ^ 2);

cSE = exp (-(nablaSW / Kappa). ^ 2);

cSW = exp (-(nablaSW / Kappa). ^ 2);

cNW = exp (-(nablaNW / Kappa). ^ 2);

% Discrete PDE solution.

im = im + delta_t * (...

(1 / (dy ^ 2)) * cN. * NablaN + (1 / dy ^ 2)) * cS. * NablaS + ...

(1 / (dx ^ 2)) * cW. * NablaW + (1 / dx ^ 2)) * cE. * NablaE + ...

(1 / (dd ^ 2)) * cNE. * NablaNE + (1 / dd ^ 2)) * cSE. * NablaSE + ...

(1 / (dd ^ 2)) * cSW. * NablaSW + (1 / dd ^ 2)) * cNW. * NablaNW + ...

end

Next, in the step (b3), an anisotropic diffusion-filtered image is binarized by the Otsu algorithm, and the Otsu algorithm automatically calculates and processes the entire image as 0 and 1 based on an arbitrary selected specific value. (See FIG. 6).

Here, the Otsu algorithm is calculated by [Equation 4] used for interclass variance, [Equation 5] and [Equation 6] used for with-class variance.

Next, in the step (b4) to remove only the throat, vertebrae, jawbone, etc. in order to select only the tooth portion, and to remove such unnecessary components must be separated from each other.

That is, in the case of teeth, they have a relatively large area because they are bounded with each other, but the others have a small area of a part, thereby reducing the image area of each object in the overall image through the erode processing of the image (see FIG. 7). .

Next, in the step (b5), only the dental information is selected as shown in the following figure by the processing of the bwareaopen function (a function used in MATLAB for image correction) to remove the remaining portions except the largest portion.

Here, as shown in FIG. 8, the partial image excluding the tooth is reduced by the erosion treatment of the image, and the portion having the largest area corresponds to the tooth portion.

Next, the step (c) may include: (c1) dilatating the auto-selected tooth image after the step (b), (c2) masking the expanded tooth image, (c3) refilling the masked tooth image, and (c4) performing morphological filtering on the infilled tooth image.

In the step (c1), as shown in FIG. 9, the selected tooth image is inflated in contrast to the erosion processed to select only the tooth information.

Next, in step (c2), as shown in FIG. 10, only the selected tooth information is extracted by masking the original CT image by using the binarization information extracted by the process of step (c1).

Next, in the step (c3), as shown in FIG. 11, the infill process is again performed to unify the portion of the tooth such as enamel, dentin, and pulp before the morphological filtering process in the tooth information. To perform.

In the end, morphological filtering treats a tooth as an object.

In addition, the step (d), (d1) extracts the edge information of the morphologically filtered tooth image after the step (c), (d2) the second polynomial fitting algorithm based on the edge information of the tooth image And extracting the dental structure by (d3) outputting the dental structure extracted through the image output apparatus.

In addition, in step (d1), as shown in FIG. 12, only the edge information of the tooth is extracted from the tooth information infill processed in step (c4) to extract the edge information of the tooth.

That is, the edge information is automatically extracted through the binarization process by the Otsu algorithm from the image as a morphologically filtered image.

Also, in the step (d2), as shown in FIGS. 13 and 14, the orthodontic structure is extracted by the second order polynomial fitting algorithm based on the edge information of the tooth.

That is, in the case of the dental information finally extracted from the CT by the described method, the enamel portion of the buccal region is thicker than the lingual side of the human tooth structure, so that the outer portion corresponding to the buccal region is dark and bright on the radiograph.

Using this advantage, the archwire used for orthodontic treatment eventually has a quadratic polynomial curve shape, and thus the orthodontic structure is extracted using the second-fit polynomial fitting algorithm.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the invention as defined in the appended claims. It will be understood that the present invention can be changed.

100: CT recording device
200: image processing apparatus
300: video output device

Claims (9)

(a) acquiring a CT image of a subject through a CT imaging apparatus;
(b) an image processing step of selecting dental information through correction and noise removal of the obtained CT image transmitted to the image processing apparatus;
(c) extracting a dental image from the imaged CT image; And
(d) outputting the dental structure of the subject in the image output apparatus through a second polynomial fitting algorithm based on the extracted tooth image;
The step (b)
(b1) Anisotropic diffusion to remove noise by differently determining the degree of filtering in each direction through an infilled CT image in which only the portion corresponding to the pulp of the tooth is erased and corrected in the selected CT image among the acquired CT images. Treating with a filter;
The anisotropic diffusion filter treatment of the step (b1),

Where I (i _f , t) is an image image at repeat step t, dvi is an divergence operator,

Is the local image gradient, C (i _f , t) is the diffusion function of the monotonically decreasing function of the image gradient size)
Compute the anisotropic process, which is the thermal conductivity equation of, to remove the noise of the dental image, and
The noise of the time (t) and the dental image is characterized in that the proportional relationship,
Automatic Extraction of Dental Structures from CT Images.
delete The method according to claim 1,
Step (b), after the step (b1),
(b2) binarizing the CT image processed by the anisotropic diffusion filter;
(b3) eroding the binarized CT image; And
(b4) selecting only dental information by automatically selecting and processing the eroded CT image;
Automatic Extraction of Dental Structures from CT Images.
The method of claim 3, wherein
The step (c)
(c1) dilatating the automated selection-treated dental image after step (b);
(c2) masking the expanded dental image;
(c3) refilling the masked tooth image; And
(c4) performing morphological filtering on the infill-treated dental image.
Automatic Extraction of Dental Structures from CT Images.
5. The method of claim 4,
The step (d)
(d1) extracting edge information of the morphologically filtered tooth image after step (c);
(d2) extracting a dental structure by the second order polynomial fitting algorithm based on the edge information of the dental image; And
(d3) outputting the extracted dental structure through the image output apparatus;
Automatic Extraction of Dental Structures from CT Images.
delete The method according to claim 1,
All images for selecting the diffusion function,

Obtained by computing the diffusion process of
The diffusion process is characterized in that the flow by the nearest four neighboring images consists of updating each pixel (x, y) in the image i _f by the same amount,
Automatic Extraction of Dental Structures from CT Images.
The method of claim 7, wherein
The four nearest neighbor images,

(only,

Is the iteration constant,

The direction of the local gradient calculated using the difference between the nearest neighbor images, and the difference between the nearest neighbor images

In the repeat step t east

, West

, Southward

, North

Means)
Characterized in that obtained by calculating the equation,
Automatic Extraction of Dental Structures from CT Images.
The method according to claim 1,
The step (a)
(a1) acquiring a plurality of dental images in a CT imaging apparatus; And
(a2) selecting one of the plurality of dental images; further comprising:
Automatic Extraction of Dental Structures from CT Images.

Images