KR101962110B1 - Method and apparatus for image segmentation - Google Patents

Abstract

According to an aspect of the present invention, there is provided an image segmentation method of an image segmentation apparatus, comprising: detecting a first region in an image based on data previously learned for a target region; Wherein the dividing step removes a part of the first area based on a parameter set in advance for the target area and a topology characteristic of the first area.

Description

[0001] METHOD AND APPARATUS FOR IMAGE SEGMENTATION [0002]

The present invention relates to image segmentation, and more particularly, to a method and apparatus for segmenting a target region in an image.

Image segmentation techniques for detecting the presence or absence of lesions in medical imaging devices such as CT (computed tomography) devices, MRI (magnetic resonance imaging) devices, and PET (positron emission tomography) devices are very important technologies. Deep learning method based on data learning is used for image segmentation in medical images. The image segmentation by the deep learning technique has a problem of low reliability and precision compared with the case where a medical staff directly reads an image.

SUMMARY OF THE INVENTION The present invention provides a method and apparatus for detecting and segmenting a target region in an image.

According to an aspect of the present invention, there is provided an image segmentation method of an image segmentation apparatus, comprising: detecting a first region in an image based on data previously learned for a target region; Wherein the dividing step removes a part of the first area based on a parameter set in advance for the target area and a topology characteristic of the first area.

Wherein the step of dividing comprises dividing the first region into a plurality of classes, labeling each class, and labeling a portion of the plurality of labeled regions with a predetermined parameter for the target region and a topology On the basis of the characteristics.

The predetermined parameter may include at least one of a position of the target region and a size of the target region.

In the removal step, the cost calculated according to the preset parameter and the cost calculated according to the topology characteristic of the first area may be used.

The predetermined parameter may include at least one of the number of pixels included in each of the labeled areas and the distance between the center of the image and each of the areas.

The topology characteristic of the first region may include at least one of an intensity, a color, an edge, a gradient, and an eigen-vector of the first region .

In the removing step, a first weight value for the cost calculated according to the preset parameter and a second weight value for the cost calculated according to the topology characteristic of the first region may be applied.

And determining whether the target region exists in the image based on the data previously learned about the target region before the detecting step.

Wherein the detecting comprises: extracting an area estimated to be the target area from the image based on data previously learned for the target area; dividing the estimated area into a plurality of areas; And removing the noise region from the region of the input signal.

The noise region may be removed using the position and size of the target region.

According to an embodiment of the present invention, there is provided an image segmentation apparatus including a detection unit detecting a first region in an image based on data previously learned about a target region, and a second region being a target region in the first region, And the dividing unit removes a part of the first area based on a parameter set in advance for the target area and a topology characteristic of the first area.

According to the embodiment of the present invention, the target area in the image can be divided. Particularly, by dividing the target area in the medical image into high-performance, it is possible to automatically determine disease and stage.

1 is a block diagram of an image segmentation apparatus according to an embodiment of the present invention.
2 is a flowchart illustrating an image segmentation method of an image segmentation apparatus according to an embodiment of the present invention.
3 is a flowchart illustrating a method of detecting a first region in an image by an image segmentation apparatus according to an exemplary embodiment of the present invention.
Fig. 4 shows the first area before noise removal.
5 shows the first area after noise removal.
6 is a flowchart illustrating a method of dividing a target region according to an embodiment of the present invention.
FIG. 7 shows the result of performing labeling for a part of the first area.
Fig. 8 shows the result of refining the image of Fig.
9 shows an image segmentation performance by the deep learning technique.
10 shows an image segmentation performance according to an embodiment of the present invention.

The present invention is capable of various modifications and various embodiments, and specific embodiments are illustrated and described in the drawings. It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention.

The terms including ordinal, such as second, first, etc., may be used to describe various elements, but the elements are not limited to these terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the second component may be referred to as a first component, and similarly, the first component may also be referred to as a second component. And / or < / RTI > includes any combination of a plurality of related listed items or any of a plurality of related listed items.

It is to be understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, . On the other hand, when an element is referred to as being "directly connected" or "directly connected" to another element, it should be understood that there are no other elements in between.

The terminology used in this application is used only to describe a specific embodiment and is not intended to limit the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In the present application, the terms "comprises" or "having" and the like are used to specify that there is a feature, a number, a step, an operation, an element, a component or a combination thereof described in the specification, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries are to be interpreted as having a meaning consistent with the contextual meaning of the related art and are to be interpreted as either ideal or overly formal in the sense of the present application Do not.

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings, wherein like or corresponding elements are denoted by the same reference numerals, and redundant description thereof will be omitted.

Although the medical image obtained from a medical imaging apparatus such as a computed tomography (CT) apparatus, a magnetic resonance imaging (MRI) apparatus, or a positron emission tomography (PET) apparatus is described in this specification, the present invention is not limited thereto. The image segmentation method and apparatus according to the embodiment of the present invention can be applied variously to fields requiring precise image segmentation such as aerial image segmentation as well as medical image segmentation. Also, the image segmentation method and apparatus according to the embodiment of the present invention can be applied not only to two-dimensional image segmentation but also to three-dimensional segmentation.

Herein, the image segmentation device may be a part of a medical image device such as a CT (computed tomography) device, an MRI (magnetic resonance imaging) device, a PET (positron emission tomography) device, Configuration.

FIG. 1 is a block diagram of an image segmentation apparatus according to an embodiment of the present invention, and FIG. 2 is a flowchart illustrating an image segmentation method of an image segmentation apparatus according to an embodiment of the present invention.

1, an image segmentation apparatus 100 includes an input unit 110, a decision unit 120, a detection unit 130, a segmentation unit 140, an output unit 150, And a storage unit 160.

Referring to FIGS. 1 and 2, an input unit 110 of the image segmentation apparatus 100 receives an image (S110). Here, the image is an image requiring segmentation of the target area, and may be, for example, a medical image. Within the medical image, the target area may be a lesion area.

The determination unit 120 determines whether there is a target area in the input image from the input unit 110 (S120). Whether there is a target area in the input image can be determined based on data previously learned about the target area. For this, the data previously learned about the target area may be stored in the storage unit 160 in advance. For example, whether there is a target area in an input image can be determined by a deep learning technique. Whether there is a target area in the input image can be represented by digital data of 0 or 1.

If it is determined that there is a target area in the image input by the determination unit 120, the detection unit 130 detects the first area in the image (S130). Here, the first area means a region estimated to be a target area, and may mean a less precisely divided area than an actual target area. For example, the first region may include a target region.

Here, the first area may be detected in the input image based on the data previously learned with respect to the target area. For this, the data previously learned about the target area may be stored in the storage unit 160 in advance. For example, the first region may be detected by a deep-running technique.

Next, the partitioning unit 140 divides the second area that is the target area in the first area (S140). At this time, the dividing unit 140 may divide the second region, which is the target region, in the first region according to a partial differential equation (PDE) principle. PDE refers to a function composed of several independent variables and an equation associated with a partial derivative of the function. When the PDE is applied to the image segmentation, a predetermined curve is set around the object to be segmented, and the set object is divided by expanding the set curve from the initial position. For this purpose, it is necessary to define the energy according to the shape of the curve and to develop the curve in the direction of minimizing it, or to change the curve repeatedly by expressing the development of the curve using the partial differential equation, You can divide the object inside the image while updating the value.

For example, according to the embodiment of the present invention, the dividing section 140 may remove a part of the first area based on a parameter set in advance for the target area and a topology characteristic of the first area. Here, a portion of the first region removed by the division unit 140 is semantically estimated to be a target region by the detection unit 130, do. Here, the preset parameter for the target area includes at least one of the position of the target area, the size of the target area, and the shape of the target area. For example, when the target region is a lesion region, at least one of the position of the lesion region, the size of the lesion region, and the shape of the lesion region may be a preset parameter. For example, lesions occurring in the brain and lesions occurring in the heart may be different in position, size, and shape in the photographed image. Accordingly, the position, size, and shape of the target region are previously set according to the type of the lesion to be segmented in the image, and may be stored in the storage unit 160 in advance. And, the topology characteristic can mean topology variation (topology derivation) or inter-pixel directionality. For example, the topology property may mean a change in at least one of intensity, color, edge, gradient, and eigen-vector.

Next, the output unit 150 displays the target area divided by the dividing unit 140 (S150). Here, the target area may be displayed with a predetermined color in the image.

Hereinafter, an image dividing method according to an embodiment of the present invention will be described more specifically.

FIG. 3 is a flowchart illustrating a method of detecting a first region in an image by an image segmentation apparatus according to an embodiment of the present invention. FIG. 4 shows a first region before noise removal, . The areas indicated in red in FIGS. 4 to 5 are the actual target areas read by the expert, and the areas indicated in green are the areas detected by the detector 130 of the image dividing device according to an embodiment of the present invention. FIGS. 4 to 5 show actual target regions together to illustrate the image segmentation performance of the image segmentation apparatus according to an embodiment of the present invention.

Referring to FIG. 3, the detection unit 130 extracts an area estimated as a target area from the input image based on the data previously learned for the target area (S300). In FIG. 4, the area extracted by step S300 is shown in green. As shown in FIG. 4, the area indicated in green is present in various places in the image.

Next, the detecting unit 130 divides the area extracted in step S300 into a plurality of areas (S310), and removes the noise area from the plurality of areas (S320). At this time, the noise region can be removed using the characteristic of the target region. For example, if the target area is a lesion area, the size and location may vary depending on the disease or organ. For example, if the region to be segmented is the heart of a human organ, the heart will be located in the center region of the image, and its size will be above a certain level. Using this feature, an area that is extracted in step S300 and does not satisfy the criterion for the feature of the target area among the plurality of areas divided in step S310 can be regarded as a noise area and can be removed. In Fig. 5, the first area after the noise area is removed is displayed.

However, the first region detected according to the method described with reference to FIG. 3 is an area that roughly includes the target area, but does not exactly match the target area. Accordingly, it is necessary to finely divide the target area from the detected first area according to the method described in Fig.

6 is a flowchart illustrating a method of dividing a target region according to an embodiment of the present invention.

Referring to FIG. 6, the dividing unit 140 divides the first region detected according to the method described in FIG. 3 into a plurality of classes (S600). The number of classes and the size of each class may vary depending on the characteristics of the target area. The number of classes may be N, for example.

Next, the partitioning unit 140 labels by class (S610). Fig. 7 (a) shows one of a plurality of divided classes, and Fig. 7 (b) shows the result of performing labeling on the class of Fig. 7 (a). Here, the labeling can be performed according to the topology characteristic. The topology property refers to the amount of topology change or directionality and can be determined according to the amount of change in at least one of intensity, color, edge, gradient and eigen-vector between pixels. have. Referring to FIG. 7 (b), there are a plurality of regions labeled with the same label, and some of them may be noise regions other than the target region.

Next, the dividing unit 140 removes a part of the plurality of labeled areas (S620). That is, the target region can be refined by removing the noise region among the plurality of regions labeled in Fig. 7 (b). Figs. 8 (a) and 8 (b) show the results of refining the images of Figs. 7 (a) and 7 (b).

At this time, the noise region, which is a part of the plurality of labeled regions, can be removed based on the parameter and the topology characteristic set in advance for the target region. For example, if the target region is a lesion region, the position and the minimum size may vary depending on the type of the lesion, and the topology characteristic may also vary. Thus, it is possible to consider a region in which the difference between the parameter and the topology characteristic set in advance for the target region out of the plurality of labeled regions out of the predetermined range is the noise region, and remove the noise region from the first region.

Here, the noise region can be removed according to the cost calculated according to the predetermined parameter with respect to the target region and the cost calculated according to the topology characteristic.

Equation (1) is an example of a cost function for removing a noise region.

Here, max (Lc) denotes the total number of pixels labeled with a predetermined value in the class, Lc (i) denotes the number of pixels in the i-th region among a plurality of regions labeled with a predetermined value in the class, and TDc (i) denotes a cost calculated according to the topology characteristic of the ith region among a plurality of regions labeled with a predetermined value in the class. W _L denotes a weight for a cost calculated according to a preset parameter for a target area, and W _t denotes a weight for a cost calculated according to a topology characteristic. W _L and W _t may be set differently depending on the type of the target area.

Equation (2) is an example of a cost function for calculating TDc (i).

는 VPE(variational problem equation)의 해이다. VPE(variational problem equation)의 해는 수학식 3으로 나타낼 수 있다.Where u is the segmented image of each labeled area and v is the original image of each labeled area. Also,

Is the solution of the variational problem equation (VPE). The solution of the VPE (variational problem equation) can be expressed by Equation (3).

은 약분으로 제거될 수 있다.Here, K is a DT (Diffusion Tensor), and? Is a value between 0 and 1 as a weight.

Can be removed with a weakening agent.

On the other hand, the topology characteristic can be represented by TD (Topologicla Derivative, D _T (x)), which can be expressed by Equation (4) below.

는 VPE(variational problem equation)의 해이고, β는 가중치로 0 내지 1 사이의 값이며, ci는 레이블링된 세그먼테이션 영상에 나타나는 클래스를 나타낸다. Where u is the segmentation image of each labeled area, v is the original image of each labeled area,

Is a solution of a variational problem equation (VPE), beta is a weight value between 0 and 1, and ci represents a class that appears in a labeled segmentation image.

At this time, steps S610 to S620 may be performed for all steps.

FIG. 9 shows an image segmentation performance by the deep learning technique, and FIG. 10 shows an image segmentation performance according to an embodiment of the present invention. 9 to 10, the red color represents the target area read by the expert, and the yellow color represents the target area divided by the image dividing device.

Referring to FIG. 9, although the area divided by the deep learning technique includes the target area read by the expert, it does not precisely divide the area. In contrast, referring to FIG. 10, it can be seen that the divided region is close to the target region read by the expert according to the embodiment of the present invention.

As used in this embodiment, the term " portion " refers to a hardware component such as software or an FPGA (field-programmable gate array) or ASIC, and 'part' performs certain roles. However, 'part' is not meant to be limited to software or hardware. &Quot; to " may be configured to reside on an addressable storage medium and may be configured to play one or more processors. Thus, by way of example, 'parts' may refer to components such as software components, object-oriented software components, class components and task components, and processes, functions, , Subroutines, segments of program code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays, and variables. The functions provided in the components and components may be further combined with a smaller number of components and components or further components and components. In addition, the components and components may be implemented to play back one or more CPUs in a device or a secure multimedia card.

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 present invention as defined by the following claims It can be understood that

Claims (12)

An image segmentation method of an image segmentation device,
Detecting a first region in the image based on previously learned data for the target region, and
And dividing a second region that is the target region within the first region,
Wherein the dividing step includes dividing the first area into a plurality of classes and labeling the plurality of classes and labeling a part of the plurality of labeled areas based on a parameter set in advance for the target area and a topology characteristic of the first area The method of claim 1, delete The method according to claim 1,
Wherein the predetermined parameter includes at least one of a position of the target region and a size of the target region. The method of claim 3,
Wherein the removing step uses the cost calculated according to the predetermined parameter and the cost calculated according to the topology characteristic of the first area. 5. The method of claim 4,
Wherein the predetermined parameter includes at least one of a number of pixels included in each of the labeled areas and a distance between the center of the image and the respective areas. 5. The method of claim 4,
Wherein the topology characteristic of the first region includes at least one of an intensity, a color, an edge, a gradient, and an eigen-vector of the first region, Way. 5. The method of claim 4,
Wherein the eliminating step applies a first weight value for a cost calculated according to the preset parameter and a second weight value for a cost calculated according to a topology characteristic of the first area. The method according to claim 1,
Before the detecting step
Determining whether the target region is present in the image based on data previously learned for the target region;
The method comprising the steps of: 9. The method of claim 8,
Wherein the detecting comprises:
Extracting an area estimated to be the target area from the image based on data previously learned for the target area;
Dividing the estimated region into a plurality of regions, and
And removing a noise region from the plurality of regions. 10. The method of claim 9,
Wherein the noise region is removed using the position and size of the target region. A detector for detecting a first area in an image based on data previously learned for a target area, and
And a dividing section for dividing the second region that is the target region within the first region,
Wherein the dividing unit divides the first area into a plurality of classes and labels each class, and divides a part of the plurality of labeled areas into a predetermined parameter for the target area and a topology characteristic of the first area Image segmentation device to remove based on. delete

Images