KR101126224B1 - Prostate segmentation method using dynamic mr images and system thereof - Google Patents

Abstract

PURPOSE: An automatic prostate partition method and a system using the same are provided to clearly partition a prostate boundary by automatically separating a prostate region from a dynamic magnetic resonance image using nasal cavity matching and subtracting techniques. CONSTITUTION: Three-dimensional volume data of prostate is acquired(S110). A magnetic resonance(MR) image is extracted by analyzing an average brightness value(S120). The MR images before/after using a cystography technique are matched using a B-spline nasal cavity matching technique(S130). The MR image in which the cystography technique is applied is acquired(S140). A partial object is extracted as a prostate candidate region(S150). An extended operation is performed in an external direction(S160). A prostate part is separated from the extracted candidate region(S170).

Description

Automatic prostate segmentation method and system using dynamic MR images

The present invention relates to an automatic prostate segmentation method and system using a dynamic MR image, and more particularly, to an automatic prostate segmentation using a dynamic MR image that automatically partitions a prostate region using non-rigid registration and subtraction in a dynamic MR image. It relates to a method and a system.

Segmenting organs from medical imaging is the most basic technique used for high-level tasks such as computer-assisted diagnosis, computer-assisted surgery, treatment planning, and biomechanical model generation. In particular, many studies have been carried out to automatically segment organs such as brain, lung, and liver from CT (Computed Tomography) images containing anatomical information and Magnetic Resonance (MR) images containing functional information. Has been. However, the prostate segmentation is still in the early stages of research because the prostate region has similar brightness to surrounding tissues and many blurry boundaries exist.

Semi-automated techniques for prostate segmentation in medical imaging began to be studied first. However, these semi-automatic segmentation techniques require a user's input process, which causes errors, resulting in a lack of reproducibility, accuracy, and objectivity of the segmentation results.

Recently, automatic techniques for prostate segmentation have been proposed. However, the existing automatic techniques are based on statistical models, and the result of segmentation of the patient data that is significantly out of the statistical model has a disadvantage in that the accuracy is greatly reduced. The problem that the generation of the statistical model greatly affects the accuracy of the segmentation There is this.

An object of the present invention is to provide an automatic prostate segmentation method and system using a dynamic MR image which automatically segmentes the prostate region using non-rigid registration and subtraction techniques.

According to the present invention, the method includes obtaining three-dimensional volume data of the prostate through a dynamic MR technique, analyzing an average brightness value from the three-dimensional volume data, and detecting a contrast-enhanced MR image. Matching the contrast-enhanced MR image with a B-spline non-rigid registration technique, subtracting the pre-contrast MR image portion from the matched contrast-enhanced MR image to obtain an imaging region MR image, and the contrast region Classifying objects through a connection element analysis technique in MR images, analyzing the positions of the centers of gravity of the objects, extracting some objects as prostate candidate regions, and expanding outwards with respect to the extracted candidate regions Automatically using a dynamic MR image comprising performing a shape propagation in an inward direction and segmenting the prostate It provides ripseon division method.

Here, the matching by the non-rigid matching technique may use a Free-Form Deformation based on the B-spline, and use Normalized Mutual Information (NMI) as a similarity measure. Can be.

The acquiring the contrast region MR image may include subtracting a partial region of the pre-contrast MR image corresponding to a lower brightness value from the contrast-enhanced MR image from the matched contrast enhancement MR image, and then subtracting the partial region. The part corresponding to the remaining area except the one can be set to zero.

The extracting of the candidate area may include extracting an area having a brightness value greater than or equal to a certain threshold value from the contrast area MR image, and then, for each slice of the contrast area MR image to divide the extracted area into the objects. Applying a connection element analysis technique, calculating an individual size and center of gravity for the slices for the objects separated by the connection element analysis technique, and a specific radius from the center of the contrast area MR image And extracting an object having the largest size among the objects having the center of gravity located in the prostate candidate region.

The extracting of the candidate region may further include filling an internal hole existing in the object having the largest size through a two-dimensional seed point region growth method.

인 원을 구성요소로 사용하는 민코우프스키 가산 방법을 통해 상기 후보 영역을 외부로 확장 연산하여, 전립선 경계점들을 생성하는 단계, 및 상기 전립선 경계점들을 제한 거리(d_max=

) 한도 내에서 내부로 형태 전파하여 전립선 경계를 분할하는 단계를 포함할 수 있다.In addition, the step of dividing the prostate, the radius size is

Extending the candidate region to the outside using a Minkowski addition method using an element as a component, generating prostate boundary points, and limiting the prostate boundary points (d _max =

Shaping the prostate border by propagating inward within limits.

The present invention also provides a computer-readable recording medium having recorded thereon a program for causing a computer to execute the automatic prostate segmentation method using the dynamic MR image.

In addition, the present invention, the acquisition unit for obtaining the three-dimensional volume data of the prostate through the dynamic MR technique, the detection unit for detecting the contrast-enhanced MR image by analyzing the average brightness value from the three-dimensional volume data, and before the contrast Matching unit for matching the MR image and the contrast-enhanced MR image by the B-spline non-rigid matching technique, and a subtractor for subtracting the pre-contrast MR image portion from the matched contrast-enhanced MR image to obtain an imaging area MR image. And an extractor for classifying objects in the contrast region MR image through a connection element analysis technique, and then extracting some objects as prostate candidate regions by analyzing positions of the centers of gravity of the objects. Dynamic MR including a divider for dividing the prostate by performing an extension operation in the outward direction and then performing a shape propagation in the inward direction An automatic prostate segmentation system using images is provided.

Here, the subtraction unit subtracts a portion of the pre-contrast MR image corresponding to a lower brightness value than the contrast-enhanced MR image from the matched contrast-enhanced MR image, and then corresponds to a portion other than the partial region. Can be set to 0.

The extractor extracts an area having a brightness value greater than or equal to a specific threshold value from the contrast area MR image, and then performs the connection element analysis technique on each slice of the contrast area MR image to divide the extracted area into the objects. Apply to each of the slices, calculate individual sizes and centers of gravity of the objects separated by the connection element analysis technique, and among the objects having a center of gravity within a specific radius from the center of the contrast region MR image. The object having the largest size may be extracted as the prostate candidate region.

인 원을 구성요소로 사용하는 민코우프스키 가산 방법을 통해 상기 후보 영역을 외부로 확장 연산하여 전립선 경계점들을 생성한 다음, 상기 전립선 경계점들을 제한 거리(d_max=

) 한도 내에서 내부로 형태 전파하여 전립선 경계를 분할할 수 있다.In addition, the division, the radius size

Using the Minkowski addition method using an element as an element, the candidate region is extended to the outside to generate prostate boundary points, and then the prostate boundary points are limited distance (d _max =

Shape propagation within limits to partition prostate boundaries.

According to the automatic prostate segmentation method and system using the dynamic MR image according to the present invention, there is an advantage that the prostate region can be automatically segmented using non-rigid registration and subtraction in the dynamic MR image. In addition, when dividing the prostate gland, there is an advantage that the prostate boundary can be clearly divided without generating leakage into adjacent tissues having similar brightness values to the prostate gland.

1 is a flowchart of an automatic prostate segmentation method using a dynamic MR image of the present invention.
2 is a schematic diagram of a system for FIG. 1.
FIG. 3 is a graph illustrating an example of analyzing average brightness values of dynamic MR images for each phase number acquired in step S110 of FIG. 1.
4 is an exemplary diagram of a subtracted image obtained through operation S140 of FIG. 1.
5 is an exemplary view of a hole filling image after analyzing connection elements through step S150 of FIG. 1.
FIG. 6 is an exemplary diagram of an externally expanded image through step S160 of FIG. 1.
FIG. 7 is an exemplary view of a prostate segmentation image through step S170 of FIG. 1.

1 is a flowchart of an automatic prostate segmentation method using a dynamic MR image of the present invention. 2 is a schematic diagram of a system for FIG. 1.

The system 100 includes an acquirer 110, a detector 120, a matcher 130, a subtractor 140, an extractor 150, and a divider 160. Hereinafter, an automatic prostate segmentation method using the dynamic MR image will be described in detail with reference to FIGS. 1 and 2.

First, the acquirer 110 acquires 3D volume data of the prostate through a dynamic MR technique (S110).

Next, the detector 120 analyzes an average brightness value from the 3D volume data to detect an 'enhanced MR image' (S120).

The dynamic MR image obtained in step S110 is taken continuously after the contrast agent is added. Therefore, the spread of the contrast medium is different from each other in the MR image of each phase photographed at different times. The first image among the acquired dynamic MR images becomes a 'pre-contrast image'.

FIG. 3 is a graph illustrating an example of analyzing average brightness values of dynamic MR images for each phase number acquired in step S110 of FIG. 1. Here, the brightness value uses HU units.

First of all, phase 1 to phase 25 have hardly been enhanced. In contrast, contrast enhancement starts from phase 25, and it can be seen that contrast enhancement is sufficient in phase 50. The MR image corresponding to phase 50 is a portion where the prostate region is brightly expressed, and can be viewed as the 'enhanced MR image'.

Next, the matching unit 130 matches the 'pre-contrast MR image' and the 'contrast-enhanced MR image' with a B-spline non-rigid matching technique (S130). Through this matching, the correspondence between the pre-contrast MR image and the contrast-enhanced MR image can be grasped.

In order to match the dynamic MR images, a global difference between the images and a local deformation due to respiration should be considered. Local deformations in dynamic MR images vary greatly from patient to patient and from the point of time of imaging. In addition, there is a problem that it is difficult to extract the same features from the MR images of each phase number due to the influence of the contrast agent.

Therefore, step S130 performs non-rigid registration using a free-form deformation model based on B-splines. The free deformation model is a method of deforming an object by moving a mesh of control points constituting a B-spline, and is suitable for modeling complex local deformation because there is no need to extract features and high degrees of freedom.

개의 제어점 격자로 구성되어 있을 때, 각 제어점

에 대하여 임의의 점

의 자유 변형

는, 1D 3차 B-스플라인으로 수학식 1과 같이 표현될 수 있다.The entire footage is

Each control point, consisting of a grid of control points

Any point against

Free deformation of

May be expressed as Equation 1 as a 1D 3rd order B-spline.

이고,

이다. 또한,

은 수학식 2와 같은 B-스플라인의 3차 기저 함수이다.here,

ego,

to be. Also,

Is the third-order basis function of the B-spline, as shown in equation (2).

The non-rigid registration technique using the B-spline deformation model has the advantage of calculating the deformation more precisely because each control point only affects the local deformation around the control point.

Here, the non-rigid registration of the 'pre-contrast MR image' and the 'contrast-enhanced MR image' is a normalized mutual information counting the number of pairs of corresponding brightness values without using the brightness value itself as a similarity measure. Normalized Mutual Information (NMI) similarity measures should be used. This is because the brightness of the prostate parenchyma and each blood vessel varies depending on the time point at which the contrast agent is injected and the time taken until the contrast agent is injected.

,

에 대한 NMI는 수학식 3으로 정의된다.Here, two images to be matched (pre-contrast MR image, contrast-enhanced MR image)

,

The NMI for is defined by Equation 3.

과

는 두 영상 각각의 엔트로피(entropy)를 나타내고,

는 두 영상의 조인트 엔트로피를 나타낸다. 각각의 엔트로피는 아래의 수학식 4로 정의된다.here,

and

Denotes the entropy of each of the two images,

Represents joint entropy of two images. Each entropy is defined by Equation 4 below.

This entropy value is calculated by creating a joint histogram. The joint histogram sets the axis to the brightness value of image 1 on the x-axis, and the brightness value of image 2 on the y-axis, and the number of brightness value pairs of image 1 and image 2 at the same position in the overlapping area of the two images. To accumulate in

After performing the non-rigid registration of the step S130 as described above, the subtractor 140 subtracts the 'pre-contrast MR image' portion from the 'matched contrast enhancement MR image' to obtain a 'contrast region MR image'. Acquire (S140).

More specifically, a portion of the 'pre-contrast MR image' corresponding to a lower brightness value than the 'contrast-enhanced MR image' is subtracted from the 'matched contrast-enhanced MR image', and then subtracted from the subtracted image. A portion corresponding to the remaining region except for the partial region is set to 0. Accordingly, only the area contrasted by the contrast agent remains. Subsequent prostate segmentation is performed on these contrasted areas.

4 shows an example of a subtracted image obtained through the step S140. Here, examples of two subtracted images (a) and (b) are shown.

After the step S140, the extractor 150 classifies the objects in the contrast area MR image, which is the subtracted image, through a connection element analysis technique, and then analyzes the positions of the centers of gravity of the objects to prostate some objects. The candidate region is extracted (S150).

Herein, the step of extracting the candidate region will be described in detail. First, an area having a brightness value greater than or equal to a specific threshold (ex, 400 HU) is extracted from the contrast area MR image acquired in step S140, and then each slice of the contrast area MR image is divided into the objects to extract the extracted area. The connection element analysis technique is applied.

The connection element analysis technique recognizes a set of pixels connected up, down, left, and right as one object, and analyzes each object by assigning a number to each connection element object separated from each other in the whole image. to be. The connection element analysis technique is well known in the art, and a detailed description thereof will be omitted.

Next, the individual size and center of gravity of the objects separated by the connection element analysis technique are calculated for each slice. To this end, among the objects whose center of gravity is located within a certain radius (ex, radius r = 20 pixels) from the center of the contrast region MR image, the object having the largest size is extracted as the prostate candidate region. This is performed under the assumption that there is a prostate candidate region within a specific radius from the center of the actually taken MR image.

In this case, since a hole may exist inside the largest connecting element object according to the noise of the MR image or the propagation state of the contrast medium, the two-dimensional seed point region grows using the outermost boundary points of the image as seed points for each slice. The method is performed to fill the inner holes existing in the object of the largest size. Since the region growth method using the seed point is a known technique, a detailed description thereof will be omitted.

An example of the hole filling image performed after the connection element analysis of step S150 is described with reference to FIG. 5. 5 (a) and 5 (b) show the results of performing step S150 on the images (a) and (b) of FIG. 4, respectively.

After extracting the candidate region, the division unit 160 performs an expansion operation in the outward direction on the candidate region extracted in step S150 (S160), and then performs shape propagation in the inward direction to divide the prostate gland ( S170). Thus, the final divided prostate border can be obtained.

FIG. 6 is an exemplary diagram of an externally expanded image through step S160 of FIG. 1. 6 (a) and 6 (b) show the results of performing step S160 on the images (a) and (b) of FIG. 5, respectively.

That is, the external expansion step S160 is to perform an expansion operation on the largest connection element found in each slice in order to detect the prostate boundary. To this end, the candidate region is extended to the outside by using the Minkowski addition method of Equation 5 using a circle having a radius size s as a component to generate prostate boundary points.

는 앞서 가장 큰 연결 요소 객체에 해당되고,

는 구성 요소(structuring element)를 나타내며,

은 실수집합에 해당된다. 그리고, 부드러운 경계 생성을 위하여 구성 요소는 반지름이

(ex,

=10픽셀)인 원 형태를 사용한다. here,

Corresponds to the largest connected element object before,

Represents a structuring element,

Is a real set. And, to create a smooth boundary, the component has a radius

(ex,

= 10 pixels).

) 한도 내에서 내부로 형태 전파하여 전립선 경계를 분할한다.In order to extract the prostate boundary from the binary image extended outward, the prostate is divided by continuously performing shape propagation inward as follows (S170). At this time, the limit distance (d _max == prostate boundary points extended through the step S160)

Shape propagates within limits to partition prostate boundaries.

That is, within the limited band value d _max , the boundary points propagate inward from the extended prostate boundary points. In order to approximate the Euclidean distance, the chess board distance of Equation 6 is used.

와

는 각각 현재 위치와 8-인접 위치를 나타낸다. 확장 경계점들에 대한

는 0으로 초기화되고, 다른 점들에 대해서는 무한대로 초기화된다. here,

Wow

Denotes the current position and the 8-adjacent position, respectively. For extended boundary points

Is initialized to zero and to infinity for other points.

인 원 형태의 구성 요소를 갖는 확장 연산을 통해 가장 큰 연결 요소를 외부로 확장시킨 후, d_max=

만큼 내부로 형태 전파를 수행하게 되면, 부드러운 전립선 경계 분할이 이루어진다. 즉, 이러한 과정을 통해 전립선 경계가 최종 분할된다.Thus, the radius

Expand the largest connected element to the outside through an extension operation with elements in the shape of circle, then d _max =

As long as the shape propagation is performed inside, smooth prostate boundary division is achieved. In other words, the prostate boundary is finally divided through this process.

FIG. 7 is an exemplary view of a prostate segmentation image result through step S170 of FIG. 1. 7 (a) and 7 (b) show the results of performing step S170 on the images (a) and (b) of FIG. 8, respectively. Referring to this, it can be seen from the segmented result that the technique of the present invention enables accurate segmentation without generating leakage into adjacent tissues having similar prostate and brightness values.

The automatic prostate segmentation method and system using the dynamic MR image of the present invention as described above, can be used as an anatomical marker when matching the CT image and MR image, area setting for prostate volume measurement and prostate detection, etc. Immediately available diagnostic tools can be provided. Furthermore, medical forms have been altered by automating and validating qualitative reading methods that depend on the manual hand of a physician now, resulting in more objective and sophisticated reading.

In addition, by bonding these systems to existing 3D MR devices or picture archive and communication systems, it can bring about significant sales and exports of software, and furthermore, competitiveness of domestic medical systems in the global market. Could be strengthened.

Meanwhile, the automatic prostate segmentation method using the dynamic MR image may be embodied as computer readable codes on a computer readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data is stored as a medium that can be read by a computer system.

Examples of computer-readable recording media include ROM, RAM, CO-ROM, magnetic tape, floppy disks, optical data storage devices, and the like, which are also implemented in the form of carrier waves (for example, transmission over the Internet). Include. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

Although the present invention has been described with reference to the embodiments shown in the drawings, these are merely exemplary and those skilled in the art will understand that various modifications and equivalent other embodiments are possible. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

100: Automatic Prostate Segmentation System Using Dynamic MR Imaging
110: acquisition unit 120: detection unit
130: matching unit 140: subtracting unit
150: extraction unit 160: division

Claims (11)

Obtaining three-dimensional volume data of the prostate through a dynamic MR technique;
Detecting an contrast enhanced MR image by analyzing an average brightness value from the 3D volume data;
Registering the pre-contrast MR image and the contrast-enhanced MR image by a B-spline non-rigid registration technique;
Acquiring a contrast region MR image by subtracting the pre-contrast MR image portion from the matched contrast enhancement MR image;
Classifying objects in the contrast region MR image through a connection element analysis technique and extracting some objects as prostate candidate regions by analyzing positions of the centers of gravity of the objects; And
And dividing the prostate by performing an extension operation in the outward direction on the extracted candidate region and then performing shape propagation in the inward direction. The method according to claim 1,
Matching with the non-rigid matching technique,
An automatic prostate segmentation method using a dynamic MR image using the Free-Form Deformation based on the B-spline and using Normalized Mutual Information (NMI) as a similarity measure. The method according to claim 1,
Acquiring the contrast region MR image,
A subtractive portion of the pre-contrast MR image corresponding to a lower brightness value than the contrast-enhanced MR image is subtracted from the matched contrast-enhanced MR image, and then set a portion corresponding to the remaining regions except for the partial region to 0. Automated Prostate Segmentation Using MR Images. The method according to claim 1,
Extracting the candidate region,
Extracting an area having a brightness value greater than or equal to a certain threshold value from the contrast area MR image, and then applying the connection element analysis technique to each slice of the contrast area MR image to divide the extracted area into the objects;
Calculating for each slice an individual size and center of gravity for the objects separated by the connection factor analysis technique; And
And extracting an object having the largest size among the objects having a center of gravity within a specific radius from the center of the contrast region MR image as the prostate candidate region. The method of claim 4,
Extracting the candidate region,
The method of claim 1, further comprising the step of filling an internal hole existing in the object having the largest size through a 2D seed point region growth method. The method of claim 4,
Dividing the prostate gland,
The radius size

Generating prostate boundary points by extending the candidate region to the outside through a Minkowski addition method using an element as a component; And
Limit the distance between the prostate boundary points (d _max =

Automatic prostate segmentation method using dynamic MR imaging comprising the step of form propagation within the limits within the boundaries of the prostate. A computer-readable recording medium having recorded thereon a program for causing a computer to execute the method according to any one of claims 1 to 6. An acquisition unit for acquiring three-dimensional volume data of the prostate through a dynamic MR technique;
A detector configured to detect an contrast enhanced MR image by analyzing an average brightness value from the 3D volume data;
A matching unit for matching the pre-contrast MR image and the contrast-enhanced MR image with a B-spline non-rigid matching technique;
A subtraction unit for subtracting the pre-contrast MR image part from the matched contrast-enhanced MR image to obtain a contrast area MR image;
An extraction unit for classifying objects in the contrast region MR image and analyzing the positions of the centers of gravity of the objects to extract some objects as prostate candidate regions; And
And a splitter configured to segment the prostate by performing an extension operation in the outward direction and then performing shape propagation in the inward direction with respect to the extracted candidate region. The method according to claim 8,
The subtraction part,
A subtractive portion of the pre-contrast MR image corresponding to a lower brightness value than the contrast-enhanced MR image is subtracted from the matched contrast-enhanced MR image, and then set a portion corresponding to the remaining regions except for the partial region to 0. Automated prostate segmentation system using MR imaging. The method according to claim 8,
The extraction unit,
Extracting an area having a brightness value greater than or equal to a specific threshold value from the contrast area MR image, and then applying the connection element analysis technique to each slice of the contrast area MR image to divide the extracted area into the objects;
Calculating for each slice an individual size and center of gravity for the objects separated by the connection factor analysis technique,
An automatic prostate segmentation system using a dynamic MR image extracting the largest object among the objects having a center of gravity within a specific radius from the center of the contrast region MR image as the prostate candidate region. The method according to claim 8,
The division part,
The radius size

Using the Minkowski addition method using an element as an element, the candidate region is extended to the outside to generate prostate boundary points, and then the prostate boundary points are limited distance (d _max =

Automated prostate segmentation system using dynamic MR imaging to shape the prostate boundary by propagating internally within limits.

Images