Segmentation-based collimation region detection and anatomical region decomposition method for X-ray images

Abstract

The present invention relates to a segmentation-based X-ray image collimation region detection and anatomical region decomposition method, which includes k-means blocks cascaded consecutively in an X-ray protocol independent manner, and morphological processing blocks used to regularize the segmentation results.

Description

[Technical Field]

[0001] The present invention relates to a segmentation-based collimation region detection and anatomical region decomposition method for X-ray images, and comprises serially cascaded k-means blocks in a manner independent of X-ray protocols, and a morphological processing block used for regularizing segmentation results. [Background Art]

[0002] Prior Art Radiography is the first imaging method dating back to the discovery of X-rays. Radiography has been widely used since the discovery of X-rays. Imaging is performed by utilizing the penetrating property of X-rays. The basic principle of radiography is to pass X-rays through all layers of the body to generate an image covering an entire region of interest.

[0003] The application of collimation regions is a management measure for limiting X-ray scattering used in radiography to reduce the radiation dose received by the patient under examination. A device that performs the collimation process in X-ray equipment is called a collimator, which consists of two sets of lead blades, each pair moving in different directions along the same axis. The collimation region refers to the area not shadowed by the lead blades, and appears as a polygon with at least four sides in an X-ray image depending on the position of the X-ray detector. However, the collimation region is not mandatory and can be omitted from the X-ray image at the discretion of the radiographer.

[0004] Segmentation is the process of generating homogeneous image objects (segments) based on features that best represent an image. In the segmentation process, image objects should correspond to real-world objects of interest. Segmentation and feature detection are fundamental to classification, and segmentation is the most important stage of object-based classification. Since the areas outside the collimation region, anatomical regions, and directly exposed regions constitute fundamentally different image classes in an X-ray image, detecting these regions using segmentation-based methods enables the development of algorithms independent of the X-ray protocol.

[0005] With the digitization of direct X-ray radiography, gains such as direct exposure through collimation and segmentation of anatomical regions are obtained on images obtained by various image processing algorithms. However, considering the workflow of radiology, the importance of algorithms that provide fast and accurate results increases. The easiest and fastest method for generating different segmentation clusters is threshold discrimination, and Patent Document 1 provides an adaptive version of threshold discrimination independently of the X-ray protocol (for imaging anatomical regions, e.g., chest X-ray). However, adaptive threshold updating methods assume that the edge transitions between direct exposure regions and anatomical regions, and between anatomical regions and collimation regions, are sharp and occur at different speeds. Candidate segmentation edge transition detection algorithms require processing each row and column of the X-ray image individually, which is highly resource-intensive considering that an average X-ray image consists of approximately 6 to 7 million pixels. Furthermore, considering the geometric shape of X-ray scattering and the thickness of the patient being imaged, it is observed that the edge transition velocities between different regions are very close to each other. On the other hand, although the method used is documented as being independent of the X-ray protocol, paragraph

[0018] of the relevant Patent Document 1 states that segmentation performance is improved when optimized according to the X-ray capture protocol, and since discrimination by adaptive thresholds is insufficient, a region expansion algorithm is used to cover defects in the segmentation mask in addition to the results of discrimination by thresholds.

[0006] As described in Patent Document 1, Patent Document 2 assumes that the transition speed and length differ between different segmentation regions in an X-ray image. Unlike discrimination using adaptive thresholds, it generates superpixels belonging to 10 different classes in an X-ray image and uses an algorithm based on edge transitions between superpixels, which is not only faster but also reflects the relationships between pixels better. However, this method assumes that transitions between different segmentation regions have different characteristics and that defects in collimation regions on superpixels are compensated for by a region expansion algorithm, which is computationally expensive.

[0007] Unlike Patent Documents 1 and 2, instead of a threshold-based discrimination method adapted to the transition between segmentation regions, the methods in Patent Documents 3 and 4 use a line-score method that employs an edge detection filter on the X-ray image. When combined with the Hough transform, this method considers the edge with the longest linearity as the collimation edge. While the accuracy of the method using the Hough transform is known to be higher than that of threshold-based discrimination algorithms, the use of this method assumes the presence of collimation regions in the X-ray image. In the absence of collimation regions, threshold-based discrimination methods can detect the absence of collimation, whereas Patent Documents 3 and 4 require an additional solution.

[0008] As a result of the relevant investigation, European Patent No. 742536 (Patent Document 5) was found. This patent application relates to recording one or more illumination regions. This patent application refers to automatically determining the location of the boundaries between multiple exposures and the boundaries between signal regions and shadow regions within each exposure. However, this patent application does not refer to the steps of image subsampling, adding virtual collimation, performing 3-center k-means segmentation, initial collimation mask detection, estimated collimation blade detection, 2-center k-means segmentation of regions outside the collimation blades, generation of morphological structural elements by adaptive dispersion threshold discrimination, completion of 3-center segmentation regions by morphological image occlusion (hidden portion repair, hole filling), and anatomical direct exposure region decomposition by 2-center k-means segmentation.

[0009] Therefore, the shortcomings and inadequacies of the existing solutions described above necessitate advancements in the relevant technological fields. [Prior art documents] [Patent Documents]

[0010] [Patent Document 1] European Patent No. 1501048 [Patent Document 2] U.S. Patent No. 5,268,967 [Patent Document 3] U.S. Patent No. 5629989 [Patent Document 4] U.S. Patent No. 5901240 [Patent Document 5] European Patent No. 742536 [Overview of the project] [Problems that the invention aims to solve]

[0011] Purpose of the invention This invention is conceived in light of the current state of the art and aims to overcome the aforementioned drawbacks.

[0012] The main objective of the present invention is to provide anatomical region resolution with fast results using extra collimation region detection independent of the X-ray capture protocol, segmentation-based X-ray image region detection, and subsampling. [Means for solving the problem]

[0013] To achieve the above-mentioned objectives, the present invention provides a segmentation-based method for detecting collimation regions and resolving anatomical regions in X-ray images, the method comprising the steps of downsampling and subsampling an X-ray image received from a detector such that the inter-pixel distribution is preserved; adding virtual collimation by virtually adding pixels of 0 in both dimensions of the downsampled X-ray image; directly exposing the downsampled X-ray image, which is known to consist of three image segments; acquiring prior information about non-collimation regions and anatomical regions and performing 3-center k-means segmentation; and merging the directly exposed segments and the anatomical region segments into a single image segment using the k-means method. The process includes the steps of creating a mask, determining four points on the mask that have the minimum Euclidean distance to the four corners of the X-ray image, generating an estimated collimation blade by connecting two consecutive points with a straight line, determining an initial collimation mask and detecting an estimated collimation blade, performing two-center k-mean segmentation of the region outside the estimated collimation blade such that the initial center values ​​are a specific proportion of the minimum pixel value and a specific proportion of the maximum pixel value of the region of interest, generating a new estimated collimation blade by re-determining the four nearest corners to the four corners of the X-ray image according to the pixel distribution of the acquired anatomical region segment and direct exposure region segment, and generating morphological structure elements by adaptive variance threshold discrimination. These processing steps include performing morphological image occlusion on the collimation mask using structural elements, completing the 3-center segmentation region by morphological image occlusion, completing the collimation segment discovery process with the original-size collimation mask, and separating the anatomically direct exposure region by 2-center k-mean segmentation.

[0014] The structural and characteristic features of the present invention and all advantages of the present invention will be more clearly understood from the drawings listed below and the detailed description given with reference to these drawings, and therefore evaluation should be made in consideration of these drawings and the detailed description.

[0015] Drawings helpful for understanding the present invention Brief Description of the Drawings

[0016] [Figure 1] It is a flowchart of the segmentation-based collimation detection and anatomical region decomposition method for X-ray images of the present invention. Mode for Carrying Out the Invention

[0017] Description of Reference Numerals 1000 Segmentation-based collimation region detection and anatomical region decomposition method for X-ray images 1001 Image sub-sampling 1002 Adding virtual collimation 1003 Three-center k-means segmentation 1004 Initial collimation mask detection and estimated collimation blade detection 1005 Two-center k-means segmentation for regions outside estimated collimation blades 1006 Morphological structuring element generation based on adaptive variance threshold discrimination 1007 Completion of three-center segmentation regions by morphological image closing (hole filling) 1008 Anatomical direct exposure region decomposition by two-center k-means segmentation

[0018] Detailed Description of the Invention In this detailed description, preferred embodiments of the novel segmentation-based X-ray image collimation region detection and anatomical region decomposition method are described solely for the purpose of better understanding the subject matter.

[0019] Images from digital X-ray imaging systems consist of three distinct segments: direct exposure, non-collimated regions, and anatomical regions. Commonly used methods in the literature for segmenting digital images are the k-means algorithm and the region expansion algorithm. Theoretically, both methods can segment the three segments in an X-ray image with a finite number of iterations, using appropriate initial segment center estimates. However, in actual testing scenarios, segmentation algorithms yield low-accuracy results due to factors such as the impact location of the X-rays, medical devices on the patient, etc. In addition, the resolution of detectors used to acquire X-ray images has increased with technological advancements; the most common adult-sized detectors currently on the market have approximately 6-7 million pixels. The iterative structure of segmentation algorithms requires reprocessing of previous pixels in the X-ray image at each iteration step, resulting in losses of processing time, processor resources, and power consumption.

[0020] The inaccurate results of segmentation algorithms are primarily due to the confusion between anatomical and non-collimated regions. When attempting to find two segment centers (collimated and non-collimated regions) instead of three, the directly exposed regions in the X-ray image are assigned to the correct segments, but the anatomical and collimated regions become mixed together. This situation creates deep gaps in the segmented image depending on the patient's position and anatomical region.

[0021] In segmented images, specific holes, gaps, or slits can be filled with appropriate structural elements through morphological image closing. However, the shapes formed as a result of segmentation by two segment centers in an X-ray image do not have a fixed geometric shape but are highly variable.

number

[0022] The segmentation-based X-ray image collimation region detection and anatomical region resolution method described above consists of k-means blocks that are sequentially connected independently of the X-ray protocol, and morphological processing blocks used to regularize the segmentation results. The mathematical expression of the k-means algorithm is given by equation (1), where k represents the number of segmentation centers, the subscript S(Si) represents the region that is adjusted by morphological processing (if applied) and segmented by the k-means algorithm, and the subscript μ(μi) represents the average pixel value of the current segmentation center. A flowchart of the segmentation-based X-ray image collimation region detection and anatomical region resolution method is shown in the figure.

[0023] Subsampling is widely used in computer vision applications to optimize resource utilization. While subsampling introduces data loss into the original image, using appropriate interpolation methods preserves most of the relationships between pixels in the original image. Since X-ray images typically consist of approximately 6 to 7 million pixels, images from an X-ray detector are subsampled at a rate of 1 in 16 using bicubic interpolation. In this way, each pixel in the subsampled image also contains information about its four neighbors in the original image.

[0024] The application of collimation is a control that limits X-ray scattering, used to reduce the amount of radiation a patient receives during X-ray imaging. The device that performs the collimation process within the X-ray machine is called a collimator and consists of two pairs of lead blades, each pair moving in different directions on the same axis. The collimation area is the part of the X-ray image that is not shadowed by the lead blades and appears as a polygon with at least four sides depending on the position of the X-ray detector. However, the collimation area is not mandatory and can be removed from the X-ray image at the discretion of the X-ray technician. To provide flexibility for different scenarios, the method identified in the diagram adds a virtual non-collimation area consisting of "0" pixels on both axes of the subsampled image.

[0025] By adding virtual collimation, a subsampled X-ray image, known to consist of three segments (direct exposure, non-collimated region, and anatomical region), is segmented using a 3-center k-means method, where the initial segment center values ​​of the 3-center k-means method are adaptively determined according to the minimum and maximum values ​​of the corresponding X-ray images. While the direct exposure region of the image acquired by segmentation can be obtained with high accuracy, the accuracy of the anatomical and non-collimated regions is very low. An initial collimation region mask is obtained by combining the direct exposure region and the anatomical region to form an initial collimation region and determining the nearest points to the four corners of the X-ray image of this region. The initial collimation region mask is a rectangle, with two opposite sides corresponding to the estimated positions of the collimation blades. The purpose of the 3-center k-means block is to acquire the direct exposure region and the anatomical region in different segments, which differ from each other in terms of pixel brightness value distribution. Using a 2-center k-means method, the anatomical and non-collimated regions can be acquired in a single segment. In addition, the information that the direct exposure area alone is responsible for the collimation area is limited.

[0026] The 3-center k-means method and the resulting estimated collimation blades underestimate the actual collimation mask. Directly exposed regions are easily separated by their high X-ray exposure, and the direct-exposed-non-collimated region boundaries and direct-exposed-anatomical region boundaries obtained by the 3-center k-means method are highly accurate. On the other hand, the transitions between anatomical regions with little or no X-ray exposure and non-collimated regions are very unclear. As a result of the 3-center k-means method, only a portion of the anatomical region can be accurately segmented. To add the missing anatomical region to the collimation region mask, the region outside the initial collimation mask is segmented by the 2-center k-means method, and the initial segmentation center values ​​of the 2-center k-means method are adaptively determined according to the minimum and maximum values ​​of the masked X-ray image. In this way, anatomical regions that were incorrectly included in the segmentation of the non-collimated region in the 3-center k-means method after the initial segmentation are detected. However, due to X-ray scattering, a small amount of X-rays may enter areas outside the collimation region, in which case some areas within the non-collimation region will approach the pixel values ​​of the anatomical region. Therefore, in the two-center k-means method, it is also necessary to decompose the blocks segmented as anatomical regions. In this relationship, the X-ray image is divided into four non-intersecting regions using a linear equation that determines the four edges of the region bounded by the estimated X-ray blade. Within each of the four regions, the two-center k-means method is checked for the presence of pixel segments found as anatomical regions. If an anatomical region block exists and the number of pixels and the variance of the segment exceed a dynamically determined threshold, the anatomical block is added to the collimation region mask obtained by the three-center k-means method.

[0027] The majority of the collimation region mask is obtained through possible additions. However, the addition of anatomical regions using the two-center k-means method does not include anatomical regions classified as non-collimated within the region formed by the estimated collimation blades of the collimation region mask. Since these gaps remain within the estimated collimation blades, it is incorrect to directly include these gaps within the collimation region because the collimation region may not be quadrilateral. Furthermore, the edges of the collimation mask must be linear, and there may be some remaining regions due to segmentation. To confine the segmentation gaps inside the estimated collimation blades and impose linearity constraints on the edges of the collimation mask, a quadrilateral region bounded by the estimated collimation blades is defined as a structural element, and morphological image closing is applied to the collimation mask. In this way, a collimation mask for the subsampled image is obtained.

[0028] The collimation mask for the subsampled image is resized to fit the size of the X-ray image using bilinear interpolation at a ratio of 1 in 16. The reason for using bilinear interpolation, unlike bicubic interpolation, in contrast to subsampling the original image, is that the 0 and 1 collimation mask generates transition bands between 0 and 1 that do not exist when enlarged, and this does not apply to bilinear interpolation. To segment the anatomical and exposed regions of the original image (if possible) by eliminating non-collimated regions, only the two-center k-means method is used, and in the two-center k-means method, the initial segmentation center value is adaptively determined by the minimum and maximum values ​​of the X-ray image masked within the collimation mask of the original size.

[0029] Segmentation-based X-ray image collimation region detection and anatomical region decomposition method; image subsampling, addition of virtual collimation, 3-center k-mean segmentation, detection of initial collimation mask and estimated collimation blade, 2-center k-mean segmentation of the region outside the estimated collimation blade, generation of morphological structure elements by adaptive dispersion threshold discrimination, completion of 3-center segmentation region by morphological image occlusion, and anatomical direct exposure region by 2-center k-mean segmentation.

[0030] In the image subsampling step, the image from the detector is reduced to a ratio of 1 in 16 pixels by bicubic interpolation, preserving the inter-pixel distribution as much as possible. This step of the above method is performed for rapid results and efficient use of processor resources.

[0031] In the virtual collimation addition step, since it is unknown whether the X-ray image with reduced pixel count is collimated or not, zero pixels are virtually added to both dimensions of the image. This step of the above method ensures that the X-ray image consists of three distinct segments, such as the direct exposure, the non-collimated region, and the anatomical region, regardless of whether collimation was applied to the original X-ray image.

[0032] In the 3-center k-means segmentation process, a reduced-pixel X-ray image, known to consist of 3 segments, is given to the k-means algorithm, the mathematical expression of which is given in equation (1). In equation (1), the number k is the number of segment centers, i.e., 3. Each segment center represents the average pixel value of the pixels that will ultimately be classified as direct exposure, non-collimation areas, or anatomical areas. The initial average pixel value of each segment center is calculated as a multiple of the minimum and maximum pixel values ​​of the hypothetical collimation image (each segment has a different coefficient). In each iteration of the k-means algorithm, the distance of 1 (1 norm) from each pixel in the image to the three segment center values, calculated in the previous iteration, is checked. Each pixel is temporarily assigned to the segment with the smallest 1 norm. The new segment center value for the next iteration is calculated from the average pixel value of the pixels temporarily assigned to each segment center in this iteration. The stopping criterion is that the segment center value remains constant for two consecutive iterations. The purpose of stopping these algorithms is to pre-generate images for direct exposure, non-collimation areas, and anatomical areas.

[0033] In the initial collimation mask detection and estimated collimation blade detection processing steps, the direct exposure segments and anatomical region segments found as a result of k-averaging are combined to form a single image mask, and four points with the minimum Euclidean distance to the four corners of the X-ray image of this mask are determined. Two consecutive points are connected by straight lines to form estimated collimation blades. This step of the above method aims to obtain positional information about the segmentation region and correct segmentation errors.

[0034] In 2-center k-means segmentation of the region outside the estimated collimation blade, each of the estimated collimation blades drawn on the X-ray image in processing step 1004 is represented by a mathematical equation of a straight line. In total, four equations of straight lines corresponding to four edges divide the X-ray image into four regions. If pixels belonging to a collimation segment exist within these four regions, the 2-center k-means algorithm is executed so that the initial center value within each region satisfying this condition is a specific percentage of the minimum pixel value and a specific percentage of the maximum pixel value in that region. The 2-center k-means algorithm in this step functions similarly to the 3-center k-means algorithm in processing step 1003, except that the number of centers is reduced to two. The purpose of this step of the above method is to fill in defects (direct exposure + anatomical regions) within the collimation region of the collimation segment obtained as a result of segmentation in processing step 1003. The 2-center k-means algorithm is used up to four times to detect possible defects in the segmentation region.

[0035] In the processing step for generating morphological structure elements by adaptive variance threshold discrimination, not all of the regions found in processing step 1005 necessarily correspond to missing collimation regions. Due to X-ray scattering, non-collimation regions may include parts that have a similar distribution to anatomical regions. To avoid adding these extra regions to the collimation region, the variance value is adaptively calculated according to the pixel distribution of the anatomical region segment and the directly exposed region segment acquired at the end of processing step 1003. Each of the possible collimation regions (up to 4 regions) outside the estimated collimation blade acquired at the end of processing step 1005 is discriminated against using a threshold corresponding to the adaptively calculated variance value, and those below the threshold are not included in the collimation mask. Regions above the threshold are merged with the initial collimation mask to form a new collimation mask. At the end of processing step 1005, after excluding possible collimation regions by adaptive variance threshold discrimination, most of the collimation mask boundary is completed. However, due to the nature of the k-means algorithm used in processing step 1003, various gaps may exist within the collimation mask. Furthermore, the possible positional scattering of the collimation region, as found by the two-center k-means algorithm used in processing step 1005, may not be physically appropriate for the linear structure of the collimation blades. To resolve these issues, after adaptive thresholding, new estimated collimation blades are generated by re-determining the four corners of the collimation mask closest to the four corners of the X-ray image. Regions bounded by these rectangular structures are filled and converted into binary structural elements. Morphological image closing processing is applied to the collimation mask using these structural elements. In this way, linearity constraints are applied to the edges of the collimation region, filling in the gaps within the collimation mask.

[0036] In the processing step to complete the 3-center segmentation region by morphological image closing, the morphological image closing process is applied to the collimation mask using the structural elements generated in processing step 1006. In this way, linearity constraints are applied to the edges of the collimation region, filling in the gaps inside the collimation mask. Subsequently, the binary image mask, which is the collimation mask, is enlarged to the original image size at a ratio of 1:16. During the enlargement of the binary image, bilinear interpolation is used in this part to avoid values ​​between pixels of 0 and pixels of 1.

[0037] In the processing steps of 2-center k-means segmentation and anatomical direct exposure region decomposition, collimation segment detection is completed with the original-size collimation mask resulting from processing step 1007. As the final step of the above method, the 2-center k-means algorithm is executed only within the collimation region mask using the k-means algorithm in equation (1). This method functions similarly to the 3-center k-means algorithm in processing step 1003, except that the number of centers is reduced to 2. Theoretically, the k-means algorithm in processing step 1003 could also find a total of 3 segments, namely: direct exposure, non-collimated region, and anatomical region. However, in actual applications (real X-ray images), anatomical and non-collimated regions are mixed together. For this reason, the task of finding collimated regions was considered as a separate, step-by-step operation. On the other hand, the decomposition of direct exposure and anatomical region by the k-means algorithm is highly accurate.

[0038] References [1] European Patent No. 1501048, title of invention "Method of segmenting a radiographic image into diagnostically relevant and diagnostically irrelevant regions" [2] U.S. Patent No. 5,268,967, title of the invention "Method for Automatic Foregrownd and Destination Detection in Digital Radiographic Images" [3] U.S. Patent No. 5,629,989, title of invention "IMAGE LINE - SEGMENT EXTRACTING APPARATUS" [4] U.S. Patent No. 5,901,240, title of the invention "Method for Detecting the Collimation Field in a Digital Radiography"

Claims

[Claim 1] A segmentation-based method for detecting collimation regions in X-ray images and for anatomical region decomposition, A processing step of downsampling the X-ray image received from the detector, A processing step of adding a virtual non-collimation region consisting of zero pixels to both dimensions of the downsampled X-ray image, and adding a virtual non-collimation region to the X-ray image from which the number of pixels has been reduced by the downsampling, A method comprising a processing step of achieving 3-center k-mean segmentation for an X-ray image with a reduced number of pixels, which is known to consist of three image segments, A processing step of determining an initial collimation mask by combining the direct exposure segment and the anatomical region segment found as a result of k-averaging into a single image mask, determining four points on the initial collimation mask that have the minimum Euclidean distance to the four corners of the X-ray image, generating estimated collimation blades by connecting the points with straight lines, and determining estimated collimation blades corresponding to the edges of the initial collimation mask, A processing step of performing two-center k-means segmentation of the region outside the estimated collimation blade such that the initial center value is a specific percentage of the minimum pixel value and a specific percentage of the maximum pixel value of each of the four non-intersecting regions formed by dividing the X-ray image using a linear equation that determines the four edges of the region bounded by the estimated collimation blade, and completing the collimation mask in the segments that exceed the threshold, A processing step of generating new estimated collimation blades by re-determining the four corners of the collimation mask closest to the four corners of the X-ray image according to the pixel distribution of the acquired anatomical region segment and the direct exposure segment, and generating binary structure elements by filling the region bounded by the new estimated collimation blades, The binary structure element is used to perform a morphological image closing process on the collimation mask, completing the three-center segmentation region by filling in the gaps in the image obscured by the morphological process. The collimation mask is enlarged to the size of the X-ray image received from the detector, the collimation segment detection process is completed using the enlarged collimation mask, and the anatomical region and the directly exposed region are separated by two-center k-averaged segmentation. A method characterized by including the following.

Images