JP2022095182A - Radiation image processing device, radiation image processing method, and program - Google Patents

Abstract

To effectively extract an area from a radiation image.SOLUTION: A radiation image processing device comprises: generation means which applies edge emphasis processing to a radiation image to generate a boundary of a target area; and extraction means which extracts the target area by using the generated boundary. The generation means generates the boundary of the target area based on either a pixel corresponding to an overshoot generated by the edge emphasis processing or a pixel corresponding to undershoot, in the radiation image.SELECTED DRAWING: Figure 1

Description

The present invention relates to a radiographic image processing apparatus, a radiographic image processing method and a program.

Conventionally, in the medical field, a radioscopic fluoroscopy apparatus that irradiates a subject with radiation, detects the radiation transmitted through the subject, and obtains a radiation image of the subject has been widely used. Radiation fluoroscopy equipment generally applies various digital image processing to radiographic images.

Among them, the area extraction process that separates the subject and the anatomical structure inside the subject from the radiographic image is a preprocessing for extracting the feature amount from the separated part, cutting out the part, and emphasizing / masking each part. As important as. As the target of region extraction, a direct irradiation region in the radiographic image in which the radiological image reaches the radiodetector directly without passing through the subject, and a lung field region of the chest radiographic image can be considered. These are areas where the amount of radiation reaching the radiation detector is relatively large. In addition, a metal region such as an implant in the subject, a bone region, and an irradiation field region where radiation reaching the radiation detector is blocked by a radiation aperture can be considered. These are areas where the amount of radiation reaching the radiation detector is relatively small. It is possible to apply it to various image processing by separating the area where the amount of radiation reaching these detectors is large or small.

For example, Patent Document 1 discloses a method of separating a direct irradiation region unnecessary for determining the gradation conversion polarity from a radiation image as a background region.

Ryo Kurazume (Author), "Level Set Method and Its Implementation Method", IPSJ Research Report, pp. 133-pp. 145, 2006.

JP-A-2014-28313

However, the method of Patent Document 1 may not always be able to effectively divide the region.

For example, when the direct irradiation line area is extracted and masked as an area requiring diagnosis or not requiring diagnosis, the area slightly above the boundary of the direct irradiation line area is not masked even to the subject area which is a diagnostic interest area. It is desirable that the boundary be placed inside.

Further, when the lung field region is extracted and cut out and displayed, it is desirable that the boundary is arranged slightly outside the boundary of the lung field region so that the cutout leakage does not occur at the end of the lung field.

The present invention has been made in view of the above problems, and one of the objects of the present invention is to effectively extract a region from a radiographic image.

It should be noted that the other object of the present invention is not limited to the above-mentioned object, but is an action and effect derived by each configuration shown in the embodiment for carrying out the invention described later, and exerts an action and effect which cannot be obtained by the conventional technique. It can be positioned as one of.

In order to solve the above object, the radiographic image processing apparatus according to the present invention applies edge enhancement processing to a radiographic image to generate a boundary of a target area, and the target using the generated boundary. The generation means includes an extraction means for extracting a region, and the generation means of the target region is based on either a pixel corresponding to an overshoot or a pixel corresponding to an undershoot generated by the edge enhancement process in the radiographic image. It is characterized by generating boundaries.

According to the present invention, a region can be effectively extracted from a radiographic image.

The figure which shows the schematic structure of the radioscopic fluoroscopy apparatus of this embodiment. A flowchart showing a processing procedure of the target area extraction unit 113 according to the first embodiment. The figure which shows an example of the sharpening filter used for the unsharp masking in the boundary generation part 114 by Example 1. The figure explaining the effect of unsharp masking in the boundary generation part 114 by Example 1. The figure explaining the case which further includes the resolution conversion unit 116 in Example 1. The figure explaining the contour model in the boundary extraction part 115 by Example 1. A flowchart showing an example of a method of creating an initial contour model in the boundary extraction unit 115 according to the first embodiment.

Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings.

[Embodiment 1]
The present invention is applied to, for example, the fluoroscopy apparatus 100 as shown in FIG. The radiation fluoroscopic imaging device 100 includes a radiation generation unit 101, a radiation aperture 102, a radiation detector 105, a data collection unit 106, a preprocessing unit 107, a CPU 108, a storage unit 109, a main memory 110, an operation unit 111, a display unit 112, and an image. A processing unit 120 and an area extraction unit 113 are provided. That is, the radioscopic fluoroscopy apparatus 100 corresponds to an example of a radiographic image processing apparatus including a generation means, an extraction means, and an image processing means. Further, each unit is connected via the CPU bus 130 and can exchange data with each other.

In the fluoroscopy apparatus 100, the main memory 110 functions as a working memory required for processing by the CPU 108. The CPU 108 uses the main memory 110 to perform operations from the operation unit 111 and control the operation of the entire device according to the parameters stored in the storage unit 109. As a result, the radiation fluoroscopy apparatus 100 operates as follows.

First, when a shooting instruction is input from the user via the operation unit 111, the shooting instruction is transmitted to the data collection unit 106 by the CPU 108. Upon receiving an imaging instruction, the data acquisition unit 106 controls the radiation generating unit 101 and the radiation detector 105 to execute radiography.

In radiography, first, the radiation generating unit 101 irradiates the subject 104 with the radiation beam 103. The radiation beam 103 irradiated from the radiation generating unit 101 is incident on the subject 104 with the irradiation range limited by the radiation diaphragm 102, passes through the subject 104 while attenuating, and reaches the radiation detector 105. Then, the radiation detector 105 outputs a signal according to the intensity of the reached radiation beam 103.

The data collection unit 106 converts the signal output from the radiation detector 105 into a predetermined digital signal and supplies it to the preprocessing unit 107 as image data.

The pre-processing unit 107 performs pre-processing such as offset correction and gain correction on the image data supplied from the data acquisition unit 106. The image data preprocessed by the preprocessing unit 107 is sequentially transferred to the main memory 110, the area extraction unit 113, and the image processing unit 120 via the CPU bus 130.

The area extraction unit 113 extracts a target area from the transferred image data and outputs the target area information.

The image processing unit 120 performs image processing on the transferred image data based on the target area information which is the result of the area extraction unit 113 extracting these target areas. This image processing includes, for example, calculation processing for calculating feature quantities such as the average value and maximum / minimum values of pixel values belonging to the target area or other than the target area, cutting / enlarging processing of only pixels belonging to the target area, and processing. It is an enhancement process such as contrast, a mask process, and the like. Further, the image processing unit 120 outputs the result of image processing based on the target area information as image processed image data. The target area information and the image processed image data are transferred to, for example, the storage unit 109 and the display unit 112 via the CPU bus 130.

The storage unit 109 stores the transferred target area information and image processed image data. The display unit 112 displays the transferred target area information and image processed image data. The user confirms the displayed target area information and the image processed image data, and gives an operation instruction as necessary via the operation unit 111.

In the radiation fluoroscopy apparatus 100 having the above configuration, the region extraction unit 113 further includes a boundary generation unit 114 and a boundary extraction unit 115.

Hereinafter, an operation in which the region extraction unit 113, which is a feature of the present embodiment, recognizes a target region from image data and outputs it as target region information will be specifically described with reference to the flowchart shown in FIG. The target region (target region) extracted by the region extraction unit 113 is a region included in the transferred image data that creates a relatively clear boundary with the soft tissue region that occupies most of the human body. be. For example, the radiation beam 103 does not pass through the subject 104, but the radiation detector 105 penetrates the direct line irradiation region reached, the lung field of the subject and the lung field region where the radiation reaches the radiation detector, and the bone portion. The bone region reached, the metal region reached through the metal, and the irradiation field region where the radiation beam 103 was blocked from reaching the radiation detector 105 by the radiation throttle 102. In other words, it is a region other than soft tissue.

(S201: Boundary generation)
In step S201, the boundary generation unit 114 included in the area extraction unit 113 obtains the result of performing the boundary generation processing of the target region to be extracted in this embodiment from the transferred input image data I0 to the boundary generation image I1. Is output as. That is, the boundary generation unit 114 corresponds to an example of a generation means that performs edge enhancement processing on a radiographic image to generate a boundary of a target area.

In this embodiment, as an example, a case where the boundary generation unit 114 outputs the result of performing unsharp masking processing on the input image data I0 as the boundary generation image I1 will be described. Unsharp masking is a process of extracting an edge portion from an image, multiplying it by a predetermined coefficient, and adding it to the original image to emphasize the edge. For example, it can be realized by spatial filtering using the sharpening filter F as shown in FIG. In the present embodiment, when the boundary generation image I1 is generated by applying the spatial filtering using the filter F to the input image data I0, it can be expressed by the following equation.

Unsharp masking will be described with reference to FIG.

It is assumed that the unsharp masking is applied to, for example, the image I0 shown in FIG. 4A to obtain the image I1 after the unsharp masking as shown in FIG. 4C. FIG. 4B shows a pixel value profile of a pixel belonging to the line indicated by the dotted line of the image I0. Further, in FIG. 4D, the pixel value profile of the pixel belonging to the line indicated by the dotted line of the image I0 is shown by a broken line, and the pixel value profile of the pixel belonging to the line indicated by the dotted line of the image I1 is shown by a solid line. That is, in FIG. 4D, the broken line shows the pixel value profile before unsharp masking, and the solid line shows the pixel value profile after unsharp masking.

In unsharp masking, as shown by the solid line in FIG. 4D, overshoot having a pixel value higher than the pixel value in the high pixel value region on the high pixel value region side across the boundary and low on the low pixel value region side. Pixel value region An undershoot with a pixel value lower than the pixel value is generated. The set of the connected pixels connecting the pixels corresponding to the overshoot or the undershoot is like the white frame and the black frame in the image I1 of FIG. 4 (c). The boundary composed of these overshoots and undershoots is a boundary intended to be generated by the boundary generation unit 114.

(S202: Boundary extraction)
Next, in step S202, the area extraction unit 113 extracts the result of extracting the boundary of the target region to be extracted in this embodiment from the boundary generation image I1 output by the boundary generation unit 114 in the boundary extraction unit 115. , Output as boundary extraction result B.

In the present embodiment, as an example, the boundary extraction unit 115 applies a method called snakes (snakes or dynamic contour model) to the boundary generation image I1 (details are in Non-Patent Document 1) to obtain a boundary. , The case of outputting as the boundary extraction result B will be described.

Snake contours a boundary on an image with a closed curve consisting of a set of pixels, defines the energy based on the continuity and smoothness of the closed curve, the pixel value at the coordinates where the closed curve is located, etc., and solves the energy minimization problem. It is a process of acquiring a desired set of boundary pixels by solving.

In the present embodiment, the energy Eshot that takes the minimum value when the closed curve is located on the overshoot or undershoot on the image I1 is defined and used as the energy of the snake. This Eshot can be expressed by, for example, the following equation.

In the number 2, N represents the number of pixels constituting the closed curve, n represents the index of the pixel, the coordinates of the nth pixel are (x (n), y (n)), and the coordinates (x (n), y ( The value on the boundary generation image I1 in (n)) is represented by I1 (x (n), y (n)). Further, (vx (n), vy (n)) represents a unit vector of normals in pixels (x (n), y (n)) (FIG. 6). That is, the following term of Equation 2 means the maximum value or the minimum value when the normal direction of the closed curve in the pixel (x (n), y (n)) is searched in the range of −M <m <M. , The maximum value is expected to be the pixel corresponding to the overshoot, and the minimum value is expected to be the pixel corresponding to the undershoot.

That is, the number 2 is an energy function that takes a minimum value of zero when the pixels (x (n), y (n)) match the maximum value or the minimum value in the respective normal directions.

Whether to use overshoot or undershoot as the boundary is selected depending on the type of image processing performed by the image processing unit 120 on the target area. Alternatively, it may be selected whether to generate the boundary of the target area based on the pixel corresponding to the overshoot or the pixel corresponding to the undershoot based on the information about the target area. The information about the target area includes, for example, at least information indicating which area in the radiographic image the target area is.

For example, when the target area is a direct irradiation area and the image processing 120 is a masking process for the direct irradiation area, the target area is made smaller so that the masked area does not erode the area of diagnostic interest. It is desirable to extract to. Here, when the direct irradiation region corresponds to the high pixel value region on the radiographic image, the target region can be extracted smaller by adopting the overshoot as the boundary.

Alternatively, when the target region is the lung field region and the image processing 120 is the cutout processing for the lung field region, the target region is extracted in a large size so that the lung field region of diagnostic interest can be extracted without any shortage at the boundary. Is desirable. Here, when the lung field region, which is the target region, corresponds to the high pixel value region on the image I1, the target region can be extracted to a large extent by adopting the undershoot as a boundary.

As described above, the boundary extraction result B extracted by the snake is output as a pixel set (x (n), y (n)) (0 ≦ n <N) constituting the closed curve. The region extraction unit 113 outputs the inside or outside of this closed curve as a target region.

In the above description, the boundary extraction unit 115 has described a method of applying a snake to the boundary generation image I1 to obtain a boundary, but the present invention can extract the boundary generated by the boundary generation unit 114. If so, it is not limited to this method. For example, there is a method called Level Set Method (Non-Patent Document 1) that uses a contour model similar to Snake, and the equation 2 may be incorporated and the boundary may be extracted using this method.

Further, the method using the contour model described above requires an initial contour model. Simply, the entire image or a small rectangle or circle in the center of the image may be used as the initial contour model, but a rough acquisition of the target area by a simple method may be given as the initial contour model.

For example, a binary image in which the input image data I0 or the boundary generation image I1 is subjected to histogram analysis to obtain a threshold value for roughly dividing the target area from the rest, and the target area is set to a pixel value of 1 and the other areas are set to a pixel value of 0 by threshold processing. To get. For this binary image, a pixel set acquired by a method such as searching for pixels selected under the condition of having a pixel with a pixel value of 1 in any of the vicinity of 4 and tracking a connected component is obtained. There is a method of using an initial contour model.

That is, the above is a method of acquiring the first boundary by threshold processing and extracting the target region from the second boundary corrected by the dynamic contour model based on the shape of at least a part of the first boundary.

Alternatively, an initial region inferior (learning model) may be generated by machine learning, and the initial contour model may be estimated using this. That is, the first boundary is estimated using a learning model generated using machine learning, and the target area is extracted from the second boundary corrected by the dynamic contour model based on the shape of at least a part of the first boundary. It may be extracted.

Here, the initial region inferior prepares a combination of a large number of radiographic images and the corresponding correct answer data of the target region in advance as supervised data, and is preferably generated by supervised learning represented by deep learning. The initial region inference device is a calculation model that calculates the initial region inference result by inputting image data using a large number of parameters generated by learning.

Since the initial region inferior generally requires a lot of calculation resources when used for an image having a large resolution, the region extraction unit 113 or the image processing unit 120 needs to extract the target region at the resolution required. Is difficult.

Therefore, in the present embodiment, a reduction process is applied to the input image data I0 or the boundary generation image I1 to generate a reduced image (step S701), and an inference process is applied to the reduced image to obtain an initial region inference result at a low resolution. Acquire (step S702). Then, the initial region inference result at low resolution is enlarged so as to have the resolution required by the region extraction unit 113 or the image processing unit 120 (step S703), and the contour is acquired by the above-mentioned method (step S704). The initial contour model is used (Fig. 7). That is, the radiation image in which the first boundary is estimated may be enlarged, and the first boundary in the enlarged radiation image may be used as the initial contour model.

By creating an initial contour model by such a method and using it in the contour extraction unit 115, more accurate boundary extraction becomes possible.

As described above, according to the present embodiment, processing for generating an overshoot or undershoot is performed, and the generated overshoot or undershoot is acquired as a boundary in a format suitable for image processing in the subsequent stage. This makes it possible to effectively extract the region from the radiographic image. In addition, it is possible to accurately extract boundaries suitable for the purpose of image processing. Further, when extracting the target region from the radiographic image, it is possible to accurately perform delicate adjustment according to the image processing to be applied at the boundary.

(Modification example)
In the above-described embodiment, the region extraction unit 113 extracts a target region based on the boundary generation image I1, and the image processing unit 120 performs image processing on the target region. Therefore, it is preferable that the boundary generation unit 114 performs processing at the image resolution desired by the region extraction unit 113 and the image processing unit 120.

That is, the boundary generation unit 114 may further include a resolution conversion unit 116 that performs resolution conversion processing such as enlargement / reduction on the input image data I0. Then, the boundary generation processing may be performed on the resolution-converted input image data I0'obtained by performing the resolution conversion processing (FIG. 5).

When the image processing unit 120 performs the boundary generation processing at a desired image resolution, it becomes possible to generate a boundary suitable for the image processing unit 120.

[Other embodiments]
The present invention supplies a program that realizes one or more functions of the above-described embodiment to a system or device via a network or storage medium, and one or more processors in the computer of the system or device reads and executes the program. It can also be realized by the processing to be performed. It can also be realized by a circuit that realizes one or more functions.

The processor or circuit may include a central processing unit (CPU), a microprocessing unit (MPU), a graphics processing unit (GPU), an application specific integrated circuit (ASIC), a field programmable gateway (FPGA). Also, the processor or circuit may include a digital signal processor (DSP), a data flow processor (DFP), or a neural processing unit (NPU).

The radioscopic fluoroscopy device in each of the above-described embodiments may be realized as a single device, or may be a form in which a plurality of devices are combined so as to be able to communicate with each other to execute the above-mentioned processing, and both of them are the embodiments of the present invention. Included in the form. The above processing may be executed by a common server device or a group of servers. The plurality of devices constituting the fluoroscopic imaging device need not be present in the same facility or in the same country as long as they can communicate at a predetermined communication rate.

In the embodiment of the present invention, a software program that realizes the functions of the above-described embodiment is supplied to the system or device, and the computer of the system or device reads and executes the code of the supplied program. Including morphology.

Therefore, in order to realize the processing according to the embodiment on the computer, the program code itself installed on the computer is also one of the embodiments of the present invention. Further, based on the instruction included in the program read by the computer, the OS or the like running on the computer performs a part or all of the actual processing, and the function of the above-described embodiment can be realized by the processing. ..

Further, the present invention is not limited to the above-described embodiment, and various modifications (including organic combinations of each embodiment) are possible based on the gist of the present invention, and these are excluded from the scope of the present invention. It's not something to do. That is, all the configurations combining the above-described embodiments are also included in the embodiments of the present invention.

101 Radiation generator 105 Radiation detector 113 Area extraction unit 114 Boundary generation unit 115 Boundary extraction unit 120 Image processing unit

Claims (18)

A generation means that applies edge enhancement processing to a radiographic image to generate a boundary of the target area,
An extraction means for extracting the target area using the generated boundary, and an extraction means.
Equipped with
The generation means is characterized in that a boundary of the target region is generated based on either a pixel corresponding to an overshoot or a pixel corresponding to an undershoot generated by the edge enhancement process in the radiation image. Processing device. The generation means creates a boundary between the region showing the soft tissue and the region other than the region showing the soft tissue.
The radiographic image processing apparatus according to claim 1, wherein the extraction means extracts a region other than the soft tissue as the target region. As the target area, the extraction means permeates the direct line irradiation region where the radiation directly reaches the radiation detector, the lung field region where the radiation reaches the radiation detector through the lung field of the subject, and the bone portion. Extract either the bone region where the radiation reaches the radiation detector, the metal region where the radiation reaches the radiation detector through the metal, or the irradiation field region where the radiation is blocked by the radiation throttle. The radiographic image processing apparatus according to claim 1 or 2, characterized in that. The radiographic image processing according to any one of claims 1 to 3, further comprising an image processing means for performing image processing on the radiographic image based on the information about the target region extracted by the extraction means. Device. The fourth aspect of claim 4, wherein the image processing means performs at least one of a feature amount calculation process, a cutout process, an enlargement process, an enhancement process, and a mask process on the radiographic image. Radiation image processing equipment. The radiation image processing apparatus according to any one of claims 1 to 5, wherein the generation means applies an unsharp masking process to the radiation image to generate a boundary of the target area. The extraction means selects whether to generate the boundary of the target area based on whether the pixel corresponding to the overshoot or the pixel corresponding to the undershoot is generated based on the information about the target area, and extracts the target area. The radiographic image processing apparatus according to any one of claims 1 to 6, characterized in that. The radiographic image processing apparatus according to claim 7, wherein the information regarding the target region includes at least information indicating which region in the radiographic image the target region is. The generation means generates a boundary based on the pixel corresponding to the overshoot when the target area is a direct irradiation area.
The radiographic image processing apparatus according to claim 7, wherein the extraction means extracts the direct irradiation region using a boundary generated based on a pixel corresponding to the overshoot. The generation means generates a boundary based on the pixel corresponding to the undershoot when the target region is the lung field region.
The radiographic image processing apparatus according to claim 7, wherein the extraction means extracts the lung field region using a boundary generated based on a pixel corresponding to the undershoot. The extraction means selects whether to generate the boundary of the target area based on whether the pixel corresponding to the overshoot or the pixel corresponding to the undershoot is generated based on the type of image processing performed by the image processing means. The radiographic image processing apparatus according to claim 5, wherein the target area is extracted. The extraction means obtains the first boundary by threshold processing, and extracts the target area from the second boundary corrected by the dynamic contour model based on the shape of at least a part of the first boundary. The radiographic image processing apparatus according to any one of claims 1 to 11. The extraction means estimates the first boundary using a learning model generated using machine learning, and the second boundary corrected by a dynamic contour model based on the shape of at least a part of the first boundary. The radiographic image processing apparatus according to any one of claims 1 to 11, wherein the target region is extracted from the subject region. The extraction means performs an enlargement process on the radiographic image in which the first boundary is estimated, and uses a dynamic contour model based on the shape of at least a part of the first boundary in the magnified radiation image. The radiographic image processing apparatus according to claim 13, wherein the target region is extracted from the corrected second boundary. The extraction means extracts the target region from the second boundary corrected by the dynamic contour model including the minimum energy function in the pixel corresponding to the overshoot or the pixel corresponding to the undershoot. The radiographic image processing apparatus according to any one of claims 12 to 14. Further provided with a resolution conversion means for converting the radiographic image to a predetermined resolution,
The generation means generates the boundary of the target area from the radiographic image whose resolution has been converted.
The radiation image processing apparatus according to claim 4, wherein the image processing means performs image processing on a radiation image whose resolution has been converted. A generation process that applies edge enhancement processing to a radiographic image to generate a boundary of the target area,
An extraction step of extracting the target area using the generated boundary, and
Equipped with
The generation step is characterized in that a boundary of the target region is generated based on either a pixel corresponding to an overshoot or a pixel corresponding to an undershoot generated by the edge enhancement process in the radiation image. Processing method. A program characterized in that a computer functions as each means of the radiographic image processing apparatus according to any one of claims 1 to 16.

Images