Medical image processing apparatus

Abstract

The present disclosure relates to a medical image processing apparatus. The medical image processing apparatus according to an embodiment of the present disclosure comprises: a memory for storing a plurality of pieces of projection data generated through detection of an X-ray; and a processor, wherein the processor can generate a first three-dimensional image for a target object on the basis of the plurality of pieces of projection data, determine a weight for the target object on the basis of the first three-dimensional image, and generate a two-dimensional image by applying the weight for the target object to a second three-dimensional image for a plurality of objects including the target object generated on the basis of the plurality of pieces of projection data.

Description

medical image processing device

[0001] The present disclosure relates to a medical image processing device.

[0002] A medical image processing device is a device that acquires and processes images of the body's internal organs. More specifically, a medical image processing device is a non-invasive examination device that captures, processes, and provides images of internal organs, fluids such as blood, and medical devices inserted into the body. Based on the medical images output from a medical image processing device, medical professionals such as doctors can assess a patient's condition and more accurately determine the presence or absence of disease.

[0003] When an X-ray imaging system captures a 2D X-ray image, the 2D X-ray image is an image in which X-rays are projected in one direction toward the X-ray imaging target. Therefore, when X-rays are projected in one direction toward the target, a problem occurs in which the image of the target overlaps and is obscured.

[0004] Meanwhile, to solve these problems, a representative device for irradiating patients with X-rays to take pictures of an object is a computed tomography (CT) device. Among medical image processing devices, a computed tomography (CT) device, which is a tomography device, can provide cross-sectional images of an object, and has the advantage of being able to express the internal structure of an object (e.g., organs such as kidneys and lungs) without overlapping compared to a general X-ray device, so it is widely used for the precise diagnosis of diseases. However, there is a problem with a computed tomography (CT) device that increases radiation exposure because it projects X-rays onto the object multiple times from various directions to take X-ray pictures.

[0005] Recently, digital tomosynthesis (DTS) X-ray imaging systems have been introduced to produce 3D images at low doses compared to computed tomography (CT). The DTS X-ray imaging system is a technique that can acquire 3D images at low doses compared to computed tomography (CT), and can provide depth-specific tomographic images to eliminate overlapping or occlusion effects.

[0006] The DTS system typically rotates an analog X-ray generator in an arch shape around a rotational axis, and the X-ray detector moves or rotates according to the position of the X-ray generator, capturing multiple X-ray images and implementing a reconstruction algorithm to create a 3D image. The 3D image created by reconstructing the X-ray images obtained through DTS can confirm the tomographic images by depth compared to 2D images, which can increase the diagnostic accuracy.

[0007] However, when only 3D images are provided, there is a disadvantage that the diagnostic load increases and the time required for diagnosis increases because the number of images to be confirmed is large. Therefore, general DTS systems create 2D images by enhancing the contrast of the region of interest learned from a large data set, creating 2D images based on averaging projection (AP), creating 2D images based on maximum intensity projection (MIP), and combining the above three methods to create 2D images. However, conventional DTS systems have a problem in that the quality of 2D images in terms of resolution or contrast is low when using existing methods. For example, in the case of digital chest tomosynthesis (DCT) for diagnosis and follow-up observation of thoracic lesions, the ribs located in the chest may obscure lesions or tissues of other organs, which may lower the diagnostic accuracy.

[0008] The present disclosure aims to solve the above-mentioned and other problems.

[0009] Another object is to provide a medical image processing device capable of providing a 2D image together with a 3D image based on an image obtained from a DTS system.

[0010] Another purpose is to provide a medical image processing device that can provide a 2D image that is easy to diagnose for objects other than a target object based on an image obtained from a DTS system.

[0011] In order to achieve the above object, a medical image processing device according to one embodiment of the present disclosure includes a memory for storing a plurality of projection data generated through detection of an X-ray; and a processor, wherein the processor generates a first three-dimensional image for a target object based on the plurality of projection data, determines a weight for the target object based on the first three-dimensional image, and applies the weight for the target object to a second three-dimensional image for a plurality of objects including the target object generated based on the plurality of projection data, thereby generating a two-dimensional image.

[0012] The effects of the medical image processing device according to the present disclosure are described as follows.

[0013] According to at least one embodiment of the present disclosure, a 2D image can be provided together with a 3D image based on an image obtained from a DTS system without additional imaging using X-rays, thereby minimizing radiation exposure of a patient.

[0014] According to at least one embodiment of the present disclosure, a 2D image can be provided together with a 3D image based on an image obtained from a DTS system, thereby improving convenience and speed of diagnosis.

[0015] According to at least one embodiment of the present disclosure, based on an image obtained from a DTS system, a 2D image that facilitates diagnosis of objects other than a target object can be provided, thereby improving the accuracy of diagnosis.

[0016] Further scope of the applicability of the present disclosure will become apparent from the detailed description below. However, since various modifications and variations within the spirit and scope of the present disclosure will become apparent to those skilled in the art, it should be understood that the detailed description and specific examples, such as preferred embodiments of the present disclosure, are given by way of example only.

[0017] Figure 1 is a drawing referenced in the description of a conventional tomosynthesis system.

[0018] FIG. 2 is a drawing for reference in the description of an X-ray imaging device according to one embodiment of the present disclosure.

[0019] FIG. 3 is a drawing for reference in an explanation of a method for capturing an X-ray image through on-off control for each of a plurality of X-ray sources according to one embodiment of the present disclosure.

[0020] FIGS. 4 and 5 are drawings for reference in an explanation of a method of taking an X-ray image using an X-ray photographing device according to one embodiment of the present disclosure.

[0021] FIG. 6 is a drawing for reference in a description of an X-ray generator in which a plurality of X-ray sources are arranged in a two-dimensional array form according to one embodiment of the present disclosure.

[0022] FIG. 7 is a diagram showing a plurality of projection data acquired by an X-ray imaging device according to one embodiment of the present disclosure.

[0023] FIG. 8 is a block diagram of an X-ray imaging device according to one embodiment of the present disclosure.

[0024] FIG. 9 is a flowchart of an operation method of an X-ray imaging device according to one embodiment of the present disclosure.

[0025] FIG. 10 is a drawing for reference in the description of a weighted projection method according to one embodiment of the present disclosure.

[0026] FIG. 11 is a diagram for comparing a tomographic image to which weighted projection is applied according to one embodiment of the present disclosure and a tomographic image reconstructed using a conventional technique.

[0027] FIG. 12 is a drawing for reference in an explanation of a method for applying a bidirectional lamp filter according to one embodiment of the present disclosure.

[0028] FIG. 13 is a diagram for comparing a tomographic image to which a bidirectional ramp filter is applied according to one embodiment of the present disclosure and a tomographic image reconstructed using a conventional technique.

[0029] FIG. 14 is a drawing for reference in a description of a method for generating a normalized tomographic image according to one embodiment of the present disclosure.

[0030] FIG. 15 is a drawing for reference in the description of an iterative reconstruction method according to one embodiment of the present disclosure.

[0031] FIG. 16 is a flowchart of an operation method of a medical image processing device according to one embodiment of the present disclosure.

[0032] FIG. 17 and FIG. 18 are drawings for reference in the description of image processing using a segmentation model according to one embodiment of the present disclosure.

[0033] FIG. 19 is a drawing for reference in a description of a plurality of bone image fragments obtained from a 3D image of a bone according to one embodiment of the present disclosure.

[0034] FIG. 20 is a drawing for reference in the description of a class map according to one embodiment of the present disclosure.

[0035] FIG. 21 is a drawing for reference in explaining fragments of an entire image according to one embodiment of the present disclosure.

[0036] FIG. 22 is a drawing for reference in the description of a 2D image according to one embodiment of the present disclosure.

[0037] Hereinafter, the present disclosure will be described in detail with reference to the drawings. In the drawings, portions irrelevant to the description are omitted to clearly and concisely describe the present disclosure, and the same reference numerals are used for identical or extremely similar portions throughout the specification.

[0038] The suffixes "module" and "part" used in the following description are given solely for the convenience of writing this specification and do not impart any particularly significant meaning or role to the components themselves. Therefore, the terms "module" and "part" may be used interchangeably.

[0039] In this application, terms such as “include” or “have” are intended to specify the presence of a feature, number, step, operation, component, part, or combination thereof described in the specification, but should be understood not to preclude the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0040] Additionally, while terms such as "first" and "second" may be used in this specification to describe various elements, these elements are not limited by these terms. These terms are used only to distinguish one element from another.

[0041] Figure 1 is a drawing for explaining a conventional tomosynthesis system.

[0042] Referring to FIG. 1, a conventional tomosynthesis system (10) irradiates an X-ray to a photographing target (102) while an X-ray generator (101) rotates about 20 to 50 degrees around a predetermined rotation axis. An X-ray detector (103) can generate an electrical signal corresponding to the dose of the X-ray transmitted.

[0043] Meanwhile, when the X-ray generator (101) rotates and irradiates the photographing target (102) with X-rays, the projected X-rays are detected by the X-ray detector (103), and a plurality of projection data (106, 107, 108) can be generated.

[0044] Meanwhile, based on the first plane (Plane 1) and the second plane (Plane 2) of the photographing target (102), each of the first point (104) and the second point (105) can be projected by the X-ray emitted by the X-ray generator (101) and mapped to each of the plurality of projection data (106, 107, 108).

[0045] In this case, the first point (104) and the second point (105) mapped to each of the plurality of projection data (106, 107, 108) may be mapped differently due to the change in the incident angle according to the rotational movement of the X-ray generator (101). Therefore, an additional step of reconstructing a 2D or 3D X-ray tomographic image of the photographing target (102) based on the plurality of projection data (106, 107, 108) is required.

[0046] FIG. 2 is a drawing for reference in the description of an X-ray imaging device according to one embodiment of the present disclosure.

[0047] Referring to FIG. 2, an X-ray photographing device (20) may include an X-ray generator (210) in which a plurality of X-ray sources (211, 212, 213, 214, 215) are arranged. The plurality of X-ray sources (211, 212, 213, 214, 215) may be turned on or off to emit X-rays, thereby irradiating an object to be photographed (220) with X-rays.

[0048] The X-ray detector (230) can generate an electrical signal corresponding to the dose of X-rays transmitted. The X-ray source can also emit X-rays in an electric field manner.

[0049] Meanwhile, the X-ray photographing device (20) may have a horizontal movement method rather than the rotational movement of the conventional tomosynthesis system (10). For example, the X-ray photographing device (20) may be controlled to emit X-rays in the order of the first X-ray source (212), the second X-ray source (213), and the third X-ray source (214).

[0050] Meanwhile, based on the first plane (Plane 1) and the second plane (Plane 2) of the shooting target (220), each of the first point (240) and the second point (250) can be mapped to each of a plurality of projection data (260, 270, 280) and shot.

[0051] In this case, the first point (240) and the second point (250) can be mapped to each of the plurality of projection data (260, 270, 280) due to horizontal movement through turning on or off each of the plurality of X-ray sources (211, 212, 213, 214, 215) of the X-ray generator (210).

[0052] Therefore, an additional step is required to reconstruct a 2D or 3D X-ray tomographic image of the photographed object (220) based on multiple X-ray images (260, 270, 280).

[0053] In this case, unlike the conventional tomosynthesis system (10) of Fig. 1, a tomographic image must be generated by reconstructing multiple projection data by reflecting the characteristics of the X-ray sources operating horizontally on or off.

[0054] FIG. 3 is a drawing for reference in an explanation of a method for capturing an X-ray image through on-off control for each of a plurality of X-ray sources according to one embodiment of the present disclosure.

[0055] Referring to FIG. 3, the X-ray photographing device (20) can take X-ray photographs of an object to be photographed in a horizontal movement manner in a first direction (u) by controlling the on or off of each of a plurality of X-ray sources included in the X-ray generator (210).

[0056] The X-ray generator (210) operates at least one X-ray source for a predetermined period of time (e.g., several msec to several hundred msec) while maintaining a gap between the X-ray sources that are turned on so that the X-ray distributions irradiated onto the X-ray detector (230) do not overlap. Meanwhile, the X-ray generator (210) can individually turn the X-ray sources on or off one by one when the X-ray distributions overlap.

[0057] The X-ray photographing device (20) can control only some of the X-ray sources to be turned on so that the X-rays emitted from the X-ray sources that are turned on do not repeatedly photograph the object to be photographed. For example, the X-ray photographing device (20) can photograph the object to be photographed by turning on the first X-ray source (311), the second X-ray source (311), and the third X-ray source (311) among the plurality of X-ray sources included in the X-ray generator (210).

[0058] After that, the X-ray photographing device (20) can turn on the fourth X-ray source (314), the fifth X-ray source (315), and the sixth X-ray source (311) to photograph the object to be photographed. In addition, the X-ray photographing device (20) can sequentially turn on the seventh X-ray source (317), the eighth X-ray source (318), and the ninth X-ray source (319) to photograph the object to be photographed.

[0059] Meanwhile, the X-ray photographing device (20) can simultaneously send a predetermined signal to the X-ray detector (230) when one or more X-ray sources are turned on, and can acquire and store projection data each time each X-ray source is turned on.

[0060] FIGS. 4 and 5 are drawings for reference in an explanation of a method of taking an X-ray image using an X-ray photographing device according to one embodiment of the present disclosure.

[0061] Referring to FIG. 4, an X-ray imaging device (20) according to one embodiment of the present disclosure can capture an X-ray image of an object (220) to be photographed while an X-ray generator (210) or an X-ray detector (230) moves horizontally in a second direction (v).

[0062] Alternatively, referring to FIG. 5, an X-ray imaging device (20) according to one embodiment of the disclosure can capture an X-ray image while the X-ray generator (210) or the X-ray detector (230) is fixed and the object to be photographed (220) moves horizontally in the second direction (v).

[0063] FIG. 6 is a drawing for reference in a description of an X-ray generator in which a plurality of X-ray sources are arranged in a two-dimensional array form according to one embodiment of the present disclosure.

[0064] Referring to FIG. 6, a plurality of X-ray sources (211) of an X-ray generator (210) can be arranged in a two-dimensional array form. The X-ray photographing device (20) can capture an X-ray image by controlling each of the plurality of X-ray sources (211) to be turned on or off in a first direction (u) or a second direction (v). Accordingly, the object to be photographed can be photographed with the same effect as when an X-ray generator in which a plurality of X-ray sources are arranged in a one-dimensional line form moves horizontally to photograph the object.

[0065] Meanwhile, FIG. 7 is a drawing showing a plurality of projection data acquired by an X-ray imaging device according to one embodiment of the present disclosure.

[0066] Referring to FIG. 7, the plurality of projection data (700) may be an X-ray projection image in which each of the plurality of X-ray sources of the X-ray generator (210) is sequentially turned on or off in a first direction (u) or the X-ray generator (210) moves horizontally in a second direction (v) and the emitted X-rays are detected and projected by the X-ray detector (230).

[0067] The imaging device (20) can reconstruct a 2D or 3D X-ray image, which is a cross-sectional image of the object to be photographed, based on a plurality of projection data (700).

[0068] Meanwhile, FIG. 8 is a block diagram of an X-ray imaging device according to one embodiment of the present disclosure.

[0069] Referring to FIG. 8, the X-ray photographing device (20) may include an X-ray generator (210), an X-ray detector (220), a memory (250), and / or a processor (270). The X-ray photographing device (20) may be referred to as a medical image processing device.

[0070] The X-ray generator (210) may include a plurality of X-ray sources. The plurality of X-ray sources may be arranged in a one-dimensional line shape or in a two-dimensional array shape.

[0071] The X-ray detector (230) can generate an electric signal corresponding to the dose of X-rays transmitted through the X-ray detector (103). The X-ray detector (230) can generate projection data by generating an electric signal.

[0072] The memory (250) stores programs for each signal processing and control within the processor (270), and can store signal-processed images, voices, or data signals, etc. The memory (250) can store a plurality of projection data.

[0073] The memory (250) may include at least one of volatile memory (e.g., DRAM, SRAM, SDRAM, etc.) or non-volatile memory (e.g., flash memory, hard disk drive (HDD), solid-state drive (SSD), etc.).

[0074] The memory (250) can store data, models, algorithms, etc. learned through machine learning such as deep learning. For example, the memory (250) can store data on the results of learning a database of medical images using X-rays through machine learning. Machine learning means that a computer learns through data without a person directly instructing the computer with logic, and thereby enables the computer to solve problems.

[0075] Deep learning is an artificial intelligence technology that teaches computers human-like thinking patterns based on artificial neural networks (ANNs), enabling computers to learn on their own like humans. ANNs can be implemented in software or hardware, such as chips. For example, ANNs can encompass various types of algorithms, including deep neural networks (DNNs), convolutional neural networks (CNNs), recurrent neural networks (RNNs), and deep belief networks (DBNs).

[0076] A pre-trained artificial intelligence algorithm model based on deep learning may be a segmentation model. For example, a segmentation model may be implemented using models such as SegNet, FCN (Fully Convolutional Network), U-net, or DeepLab. In this disclosure, the segmentation model is implemented using U-net as an example, but is not limited thereto.

[0077] The processor (270) can control the movement of the X-ray generator (210) or the X-ray detector (230), or control the on or off of each of the multiple X-ray sources of the X-ray generator (210).

[0078] Additionally, the processor (270) can store a plurality of projection data generated from the X-ray detector (230) in the memory (250).

[0079] Additionally, the processor (270) can reconstruct a plurality of projection data into a 2D or 3D X-ray image, which is a cross-sectional image of the photographing target. For example, the processor (270) can apply a predetermined reconstruction algorithm based on a plurality of projection data to generate a 2D or 3D X-ray cross-sectional image.

[0080] Meanwhile, a representative reconstruction algorithm is FBP (filtered back projection). However, when the FBP (filtered back projection) reconstruction algorithm used in the conventional tomosynthesis system (10) disclosed in FIG. 1 is used in the X-ray photographing device (20) of FIG. 2, a problem of artifacts occurring in the reconstructed X-ray image may occur. In the conventional tomosynthesis system (10), the X-ray generator rotates around a predetermined rotation axis to obtain projection data, but the X-ray photographing device (20) of FIG. 2 moves horizontally to obtain projection data. Therefore, since the X-rays are incident at a limited angle on a part of the X-ray detector, only a part of the projection data is used to calculate one pixel value, and thus, discontinuous linear image artifacts may appear in the reconstructed X-ray image.

[0081] The processor (270) can detect at least one anatomical structure included in a medical image. The processor (270) can detect at least one of a plurality of anatomical structures included in the medical image using a deep learning-based artificial intelligence algorithm learned through machine learning. Here, the deep learning-based artificial intelligence algorithm learned through machine learning may be a model stored in the memory (250). For example, the processor (270) can detect a rib from an image corresponding to a chest using a segmentation model stored in the memory (250).

[0082] The processor (270) may include a data learning unit (not shown) that learns medical images through machine learning such as deep learning, and a data processing unit (not shown) that processes medical images using an artificial intelligence algorithm based on deep learning learned through machine learning.

[0083] FIG. 9 is a flowchart of an operation method of an X-ray imaging device according to one embodiment of the present disclosure.

[0084] Referring to FIG. 9, the processor (270) can obtain multiple projection data (S901). For example, the processor (270) can obtain multiple projection data generated by the X-ray detector (230) and stored in the memory (250).

[0085] The processor (270) may perform log projection on each of the plurality of projection data to generate log-projected projection data (S902). Log projection may change the exponential characteristics of the projection data reflected due to the exponential absorption of X-rays by the photographing subject (220) into a log scale so that the projection data has a linear value.

[0086] Additionally, the processor (270) can perform weighted projection on log-projected projection data to generate weighted projection data (S903).

[0087] Referring to the plurality of projection data (700) of FIG. 7, each projection data creates a boundary of an X-ray irradiation area for a portion of a photographing target. Therefore, when the projection data is backprojected to the FOV (Field Of View) area corresponding to the projection data, artifacts may occur near the boundary. Therefore, the processor (270) performs weighted projection on the log-projected projection data to generate weighted projection data, thereby reducing artifacts occurring near the boundary.

[0088] FIG. 10 is a drawing for reference in the description of a weighted projection method according to one embodiment of the present disclosure.

[0089] Referring to FIG. 10, the processor (270) projects log-projected projection data ( ) with weights ( ) to apply the weighted projection data ( ) can be generated. The weighted projection can be applied to the following mathematical formula 1.

[0090]

[0091]

[0092] Here, u and v can be the coordinates of the X-ray source.

[0093] FIG. 11 is a diagram for comparing a tomographic image to which weighted projection is applied according to one embodiment of the present disclosure and a tomographic image reconstructed using a conventional technique.

[0094] Referring to FIG. 11, when comparing a tomographic image (1101) reconstructed using a conventional technique with a tomographic image (1102) to which weighted projection is applied, it is possible to confirm the effect of reducing artifacts occurring near the boundary line.

[0095] Meanwhile, the processor (270) weighted projected projection data ( ) can be used to generate projection data with a two-way ramp filter applied (S904).

[0096] FIG. 12 is a drawing for reference in an explanation of a method for applying a bidirectional lamp filter according to one embodiment of the present disclosure.

[0097] A ramp filter may be a filter that emphasizes high frequency components in a certain direction to compensate for the asymmetric appearance of frequency elements depending on the acquisition direction of projection data. Accordingly, it can produce the effect of making objects appear clearer.

[0098] An image capturing device (20) according to one embodiment of the present disclosure sequentially irradiates an object to be captured with X-rays in a first direction (e.g., u direction) and a second direction (e.g., v direction) to obtain projection data. Since the object is captured by moving horizontally in the first direction (u) or the second direction (v), an area in which the object to be captured disappears occurs in each of a plurality of projection data.

[0099] For example, when the X-ray generator (210) includes a plurality of X-ray sources arranged in a one-dimensional line shape, each of the plurality of X-ray sources is sequentially turned on to irradiate X-rays to the photographing target in a first direction. In addition, the X-ray generator (210) can move in a second direction to take photographs.

[0100] Therefore, when a ramp filter (1D ramp filter) is applied in only one direction for each of multiple projection data, objects in the projection data may be blurred in only one direction, resulting in an asymmetrical shape.

[0101] Therefore, the processor (270) can apply a ramp filter in both directions (first direction and second direction).

[0102] Referring to FIG. 12, the processor (270) weighted projected projection data ( ) projection data with a ramp filter applied in the first direction (u) ) can be obtained. In addition, the processor (270) can obtain weighted projected projection data ( ) projection data with a ramp filter applied in the second direction (v) ( ) can be obtained. In addition, the processor (270) can obtain weighted projected projection data ( ) projection data with a bidirectional ramp filter applied to it ( ) can be obtained. Projection data with a lamp filter applied in both directions ( ) can be applied to the following mathematical formula.

[0103]

[0104] FIG. 13 is a diagram for comparing a tomographic image to which a bidirectional ramp filter is applied according to one embodiment of the present disclosure and a tomographic image reconstructed using a conventional technique.

[0105] Referring to Fig. 13, when comparing a tomographic image (1301) reconstructed using a conventional technique with a tomographic image (1302) to which a bidirectional ramp filter is applied, it can be confirmed that the phenomenon of objects being blurred in only one direction and thus creating an asymmetrical shape has been reduced.

[0106] Meanwhile, the processor (270) can generate a reconstructed single-layer image based on each projection data to which a bidirectional ramp filter is applied (S905). In this case, the image reconstruction step in a general filtered back projection (FBP) algorithm may be applied.

[0107] Additionally, the processor (270) can apply normalization to the reconstructed tomographic image to generate a normalized reconstructed tomographic image (S906).

[0108] FIG. 14 is a drawing for reference in a description of a method for generating a normalized tomographic image according to one embodiment of the present disclosure.

[0109] Referring to Fig. 14, the plurality of projection data is data captured by a plurality of X-ray sources moving horizontally in the first or second direction. Accordingly, as one moves from the edge of the photographing object to the center of the photographing object, more X-rays can be irradiated, and as one moves toward the center of the photographing object, more pixels overlap in the projection data.

[0110] Therefore, the processor (270) reconstructs the weights used in the weighted projection to generate a sampling density map (SD) and generates a reconstructed tomographic image ( ), the normalized reconstructed tomographic image ( ) can be generated. Reconstructed single-layer images ( ), sampled density map (SD), normalized reconstructed tomographic image ( ) can be applied to the following mathematical formula.

[0111]

[0112]

[0113]

[0114]

[0115] FIG. 15 is a drawing for reference in the description of an iterative reconstruction method according to one embodiment of the present disclosure.

[0116] Referring to FIG. 15, the processor (270) may repeatedly perform the reconstruction process after initially acquiring a normalized reconstructed tomographic image. In this case, the processor (270) may utilize a maximum likelihood-expectation maximization (ML-EM) algorithm, etc.

[0117] The processor (270) can repeat the iterative reconstruction process on the normalized reconstructed tomographic image until an iteration criterion is satisfied. The iteration criterion can be set to a predetermined number of iterations.

[0118] The processor (270) can acquire the first generated normalized reconstructed tomographic image as the first tomographic image (S1501).

[0119] The processor (270) can obtain a plurality of virtual projection data by forward projecting the first single-layer image based on the coordinates (u, v) of each of the plurality of X-ray sources (S1502).

[0120] The processor (270) can obtain a difference or ratio by comparing each of the plurality of projection data and the plurality of virtual projection data obtained by irradiating the photographing object with X-rays based on the coordinates of each X-ray source (S1503).

[0121] The processor (270) can obtain a back-projected image difference or ratio by applying weighted projection to the obtained difference or ratio and performing back-projection (S1504).

[0122] Additionally, the processor (270) can obtain a second tomographic image by applying normalization based on the backprojected image difference or ratio and the first tomographic image (S1504).

[0123] Additionally, the processor (270) can determine whether further repetitions are required (S1506). For example, if a predetermined number of repetitions has been met, the processor (270) can determine that further repetitions are not required and terminate the iterative reconstruction (S1507). Furthermore, for example, if the predetermined number of repetitions has not been met, the processor (270) can determine that further repetitions are required and perform the reconstruction process again on the second tomographic image (S1502).

[0124] Meanwhile, the n+1 tomographic image that has undergone the iterative reconstruction process for the nth tomographic image can satisfy the following mathematical equation.

[0125]

[0126] Fig. 16 is a flowchart illustrating an operating method of a medical image processing device according to an embodiment of the present disclosure. Any details that overlap with those described in Figs. 1 to 15 will be omitted for brevity.

[0127] Referring to FIG. 16, the medical image processing device (20) can, in operation S1610, perform segmentation on a target object included in each of a plurality of projection data. In the present disclosure, a bone among various organs located inside the body is set as a target object, but the present disclosure is not limited thereto.

[0128] The processor (270) of the medical image processing device (20) can perform segmentation on bones included in each of the plurality of projection data using a pre-learned segmentation model stored in the memory (250). Hereinafter, the description will be based on the ribs located in the chest, but is not limited thereto.

[0129] Referring to FIG. 17, a segmentation model according to one embodiment of the present disclosure may be a U-net, which is an encoder-decoder based model having multiple convolution layers.

[0130] U-net may be a network that encodes projection data (1710) through a convolution layer, decodes it through a transposed convolution layer, and obtains segmentation masking as the result data (1720). The transposed convolution layer may also be called a deconvolution layer or an upconvolution layer. The encoder and decoder may use 3x3 convolution.

[0131] In one embodiment, the loss functions used in U-net may include Binary cross entropy (BCE), dice coefficient (DICE), and / or Total Variation (TV) loss.

[0132] The medical image processing device (20) can generate a 3D image of a bone based on a plurality of result data obtained by performing segmentation on the bone, each corresponding to a plurality of projection data, in operation S1620.

[0133] Referring to FIG. 18, a medical image processing device (20) can obtain a plurality of result data (1800) obtained by performing segmentation on a bone, each corresponding to a plurality of projection data.

[0134] The medical image processing device (20) can perform log projection on each of the plurality of result data (1800) to generate log-projected result data. The medical image processing device (20) can perform weighted projection on the log-projected result data to generate weighted projected result data. The medical image processing device (20) can apply a bidirectional ramp filter to the weighted projected result data to generate result data to which the bidirectional ramp filter is applied. The medical image processing device (20) can generate a 3D image of a reconstructed bone based on each of the result data to which the bidirectional ramp filter is applied. In this case, an image reconstruction step in a general filtered back projection (FBP) algorithm may be applied. Hereinafter, the 3D image of the bone may be referred to as a 3D bone mask.

[0135] Referring to FIG. 19, a 3D bone mask may be composed of a plurality of image slices (hereinafter, bone image slices) sliced ​​in the depth direction. In the present disclosure, the depth direction may be the z-axis direction. That is, each of the plurality of bone image slices may be an image slice in which the 3D bone mask is sliced ​​along the z-axis direction. For example, when the 3D bone mask is sliced ​​into 50 image slices along the z-axis direction, the medical image processing device (20) may acquire 50 bone image slices. At this time, the index value of the z-axis may have a value from 0 to 49 or from 1 to 50. That is, the 3D bone mask may be a stack of 50 bone image slices.

[0136] The medical image processing device (20) may, in operation S1630, determine a plurality of classes corresponding to the positions of bones included in the 3D bone mask. For example, the plurality of classes may include a first class corresponding to a rib portion located on the front side of the chest (hereinafter, “front rib”), a second class corresponding to a rib portion located on the back side of the chest (hereinafter, “back rib”), and a third class corresponding to an uncertainty portion. Meanwhile, an area that does not correspond to the first to third classes may be determined as a fourth class corresponding to the background.

[0137] According to one embodiment, the medical image processing device (20) can determine a class corresponding to the location of a bone through K-mean clustering. Here, K-mean clustering may refer to a clustering technique that selects a specific arbitrary point to set a cluster center point and selects points closest to the center. For example, the medical image processing device (20) can apply K-mean clustering in the depth direction to a 3D bone mask to classify a bone area included in the 3D bone mask into a plurality of classes. At this time, when the processor (270) classifies the bone area included in the 3D bone mask into three classes, it can set three cluster center points corresponding to each of the three classes.

[0138] The medical image processing device (20) can generate a class map based on the class determined through K-mean clustering in operation S1640. Here, the class map can be divided into a plurality of classes corresponding to each part of the rib. For example, the processor (270) can extract indices with the largest pixel value in the depth direction for the 3D bone mask. At this time, the processor (270) can generate a class map using the extracted indices and the pixel value of each index. The class map can represent 3D spatial data. The class map can be a manifold representing 3D spatial data.

[0139] Referring to FIG. 20, the class map (2000) may include a front lip (2010) corresponding to a first class, a back lip (2020) corresponding to a second class, and an uncertain part (2030) corresponding to a third class.

[0140] The medical image processing device (20) can classify a plurality of index values ​​based on the location of the bone according to the class map in operation S1650.

[0141] According to one embodiment, the medical image processing device (20) may calculate the average and standard deviation of the index values ​​in the depth direction for each of a plurality of classes. For example, when the index values ​​of the z-axis have values ​​from 1 to 50, the average calculated for the first class may be 10, and the average calculated for the third class may be 43. The average calculated for the second class may be greater than 10 and less than 43. Meanwhile, the standard deviation calculated for the second class may be greater than the standard deviation calculated for the first class or the third class.

[0142] The medical image processing device (20) can classify a plurality of index values ​​based on the average and standard deviation calculated for each of a plurality of classes. For example, among the index values ​​of the z-axis, the index values ​​corresponding to 1 to 20 can be classified into a first index set in which the proportion of the front lip is large, the index values ​​corresponding to 21 to 35 can be classified into a second index set in which the proportion of the rib is small, and the index values ​​corresponding to 36 to 50 can be classified into a third index set in which the proportion of the back lip is large.

[0143] The medical image processing device (20) may, in operation S1660, acquire multiple full image slices from a full 3D image of multiple objects. Here, the multiple objects may include ribs, lungs, a heart, etc. For example, the full 3D image may be a reconstructed tomographic image acquired according to the embodiment of FIG. 9. For example, the full 3D image may be a second tomographic image acquired according to the embodiment of FIG. 15.

[0144] Each of the entire image fragments may be a 2D image. The medical image processing device (20) may acquire a plurality of entire image fragments by dividing the entire 3D image in the depth direction. That is, each of the plurality of entire image fragments may be an image fragment in which the entire 3D image is divided along the z-axis direction. The number of the plurality of entire image fragments may correspond to an index value for the 3D bone mask. For example, if the index value for the 3D bone mask is 50, the entire 3D image may be a stack of 50 entire image fragments.

[0145] Referring to FIG. 21, in each of a plurality of full image fragments (2100), the proportion occupied by the ribs may be different. In the case of the first image fragment (2110) having a relatively small z-axis index value, since the proportion occupied by the front lip is large, other organs may be obscured by the front lip. In addition, in the case of the third image fragment (2130) having a relatively large z-axis index value, since the proportion occupied by the back lip is large, other organs may be obscured by the back lip. Meanwhile, in the case of the second image fragment (2120) having a z-axis index value between the index value of the first image fragment (2110) and the index value of the third image fragment (2130), since the proportion occupied by the ribs is small, organs such as the lungs excluding the ribs may be confirmed as a whole. In this way, since the proportion of the ribs in the entire image fragment varies depending on the index value of the z-axis, when a 2D image is created using only multiple entire image fragments, organs such as the lungs located in the chest may be obscured by the ribs in the 2D image.

[0146] The medical image processing device (20) can generate a 2D image based on weights corresponding to each of a plurality of index values ​​in operation S1670. The medical image processing device (20) according to one embodiment can generate a 2D image by performing weighted summation based on the following mathematical equations.

[0147]

[0148]

[0149]

[0150] Here, E represents a weighted mask according to bone proportion, A represents a set of index values ​​of the z-axis with a large bone proportion, W represents a 3D bone mask, λ represents a trade-off parameter, I represents an identity matrix, f represents the entire 3D image, and S represents a 2D image with a reduced bone proportion. Meanwhile, Gaussian smoothing can be applied to E.

[0151] Based on the above mathematical formulas, since 1-W is multiplied to all image fragments corresponding to index values ​​with a large proportion of ribs (e.g., 1 to 20, 36 to 50), the proportion of all image fragments with a large proportion of ribs can be reduced when generating a 2D image. In addition, all image fragments corresponding to index values ​​with a small proportion of ribs (e.g., 21 to 35) Since λ is multiplied, entire image fragments with small rib proportions can be applied as is when generating a 2D image. At this time, the larger λ is, the more a 2D image is generated that is concentrated on index values ​​with small rib proportions, and the smaller λ is, the more a 2D image can be generated to which multiple index values ​​are uniformly applied.

[0152] Meanwhile, the medical image processing device (20) can also generate a 2D image by applying various methods such as Maximum Intensity Projection (MIP) and Average Intensity Projection (AIP).

[0153] FIG. 22 is a drawing for reference in a description of a 2D image generated according to one embodiment of the present disclosure.

[0154] Referring to FIG. 22, in the case of a 2D image (2210) generated by simply synthesizing entire image fragments without considering the proportion of bones, other organs located in the chest may be obscured as the ribs are displayed as they are. In contrast, in the case of a 2D image (2220) generated according to an embodiment of the present disclosure, the proportions of entire image fragments with a large proportion of ribs are small, and the proportions of entire image fragments with a small proportion of ribs are large, so that the degree to which other organs located in the chest are obscured by the ribs can be minimized.

[0155] Meanwhile, the present disclosure is described based on images acquired through a DTS X-ray imaging system, but is not limited thereto. For example, the medical image processing device (20) can generate a 2D image based on weights corresponding to each of a plurality of index values ​​from a 3D image acquired through computed tomography (CT).

[0156] As described above, according to at least one embodiment of the present disclosure, a 2D image can be provided together with a 3D image based on an image obtained from a DTS system without additional imaging using X-rays, thereby minimizing radiation exposure of a patient.

[0157] In addition, according to at least one embodiment of the present disclosure, a 2D image can be provided together with a 3D image based on an image obtained from a DTS system, thereby improving convenience and speed of diagnosis.

[0158] In addition, according to at least one embodiment of the present disclosure, based on an image obtained from a DTS system, a 2D image that facilitates diagnosis of objects other than the target object can be provided, thereby improving the accuracy of diagnosis.

[0159] Referring to FIGS. 1 to 22, a medical image processing device (20) according to one aspect of the present disclosure includes a memory for storing a plurality of projection data generated through detection of an X-ray; and a processor, wherein the processor generates a first three-dimensional image for a target object based on the plurality of projection data, determines a weight for the target object based on the first three-dimensional image, and applies the weight for the target object to a second three-dimensional image for a plurality of objects including the target object generated based on the plurality of projection data, thereby generating a two-dimensional image.

[0160] In addition, according to one aspect of the present disclosure, the memory stores a pre-learned deep learning-based artificial intelligence algorithm, and the processor performs segmentation of the target object for each of the plurality of projection data using the artificial intelligence algorithm, and generates the first 3D image based on a plurality of result data of performing the segmentation corresponding to each of the plurality of projection data.

[0161] Additionally, according to one aspect of the present disclosure, the artificial intelligence algorithm may be a U-net, which is an encoder-decoder based model having multiple convolutional layers.

[0162] Additionally, according to one aspect of the present disclosure, the plurality of objects may be included in an organ, and the target object may be a bone.

[0163] In addition, according to one aspect of the present disclosure, the processor can generate a class map that divides areas of the target object included in the first 3D image into a plurality of classes, and determine a weight for the target object based on the class map.

[0164] In addition, according to one aspect of the present disclosure, the processor may classify, based on the class map, a plurality of index values ​​for a predetermined direction into a first index set corresponding to the target object and a second index set not corresponding to the target object, and determine a first weight corresponding to the first index set and a second weight corresponding to the second index set as weights for the target object.

[0165] Additionally, according to one aspect of the present disclosure, the processor can classify regions of the target object included in the first 3D image into the plurality of classes through K-mean clustering.

[0166] Additionally, according to one aspect of the present disclosure, the processor may extract index values ​​having the largest pixel value in a predetermined direction, and generate the class map based on the extracted index values ​​and the pixel value corresponding to each of the extracted index values.

[0167] Additionally, according to one aspect of the present disclosure, the plurality of classes may include a first class corresponding to a first portion located on the side of the X-ray generator; a second class corresponding to a second portion located on the rear side of the body; and a third class corresponding to an uncertainty portion.

[0168] In addition, according to one aspect of the present disclosure, the processor may calculate, for each of the plurality of classes, an average and a standard deviation of index values ​​for the predetermined direction, and classify the plurality of index values ​​into the first index set and the second index set based on the average and the standard deviation calculated for each of the plurality of classes.

[0169] In addition, according to one aspect of the present disclosure, the plurality of projection data are generated based on detection of the X-rays irradiated from the plurality of X-ray sources, and the processor can apply log projection to each of the plurality of result data, apply weighted projection to the log-projected result data, apply a bidirectional ramp filter to the weighted projected result data, and generate the first 3D image reconstructed based on each of the result data to which the bidirectional ramp filter is applied.

[0170] In addition, according to one aspect of the present disclosure, the processor may obtain a plurality of image pieces sliced ​​along a predetermined direction from the second three-dimensional image, and apply one of the first weight and the second weight to each of the plurality of image pieces to generate the two-dimensional image.

[0171] Additionally, according to one aspect of the present disclosure, the processor may generate the two-dimensional image by performing a weighted summation, applying the first weight and one of the above, to each of the plurality of image fragments.

[0172] In addition, according to one aspect of the present disclosure, the processor calculates the first weight and the second weight based on the following mathematical expression 1,

[0173] [Mathematical Formula 1]

[0174]

[0175] The above E may be a weighted mask according to the specific gravity of the target object, the above A may be a first index set, the above W may be a mask corresponding to the first 3D image, λ may be a trade-off parameter, and I may be an identity matrix.

[0176] In addition, according to one aspect of the present disclosure, the processor performs the weighted summation based on the following mathematical expressions 2 and 3,

[0177] [Equation 2]

[0178]

[0179] [Equation 3]

[0180]

[0181] The above f may be the second 3D image, and the above S may be the 2D image.

[0182] The attached drawings are only intended to facilitate understanding of the embodiments disclosed in this specification, and the technical ideas disclosed in this specification are not limited by the attached drawings, and should be understood to include all modifications, equivalents, or substitutes included in the spirit and technical scope of the present disclosure.

[0183] Meanwhile, the operating method of the present disclosure can be implemented as processor-readable code on a processor-readable recording medium. A processor-readable recording medium includes all types of recording devices that store data that can be read by a processor. Examples of processor-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage devices, etc., and also include those implemented in the form of a carrier wave, such as transmission via the Internet. Furthermore, the processor-readable recording medium can be distributed across network-connected computer systems, so that the processor-readable code can be stored and executed in a distributed manner.

[0184] In addition, although the preferred embodiments of the present disclosure have been illustrated and described above, the present disclosure is not limited to the specific embodiments described above, and various modifications may be made by a person having ordinary skill in the art to which the present invention pertains without departing from the gist of the present disclosure as claimed in the claims, and such modifications should not be understood individually from the technical idea or prospect of the present disclosure.

Claims

1. A memory for storing multiple projection data generated through detection of X-rays; and Contains a processor, The above processor, Based on the above plurality of projection data, a first three-dimensional image of the target object is generated, Based on the first 3D image, a weight for the target object is determined, A medical image processing device characterized in that it generates a two-dimensional image by applying a weight for the target object to a second three-dimensional image for a plurality of objects including the target object generated based on the plurality of projection data.

2. In paragraph 1, The above memory stores a pre-learned deep learning-based artificial intelligence algorithm, The above processor, Using the above artificial intelligence algorithm, segmentation of the target object is performed for each of the plurality of projection data, A medical image processing device characterized in that it generates the first three-dimensional image based on a plurality of result data corresponding to each of the plurality of projection data and performing the segmentation.

3. In paragraph 2, The above artificial intelligence algorithm, A medical image processing device characterized by being a U-net, an encoder-decoder based model having multiple convolutional layers.

4. In paragraph 1, The above multiple objects are included in an organ, A medical image processing device, characterized in that the target object is a bone.

5. In paragraph 1, The above processor, Generate a class map that divides areas of the target object included in the first 3D image into multiple classes, A medical image processing device characterized in that a weight for the target object is determined based on the class map.

6. In paragraph 5, The above processor, Based on the above class map, a plurality of index values ​​for a given direction are classified into a first index set corresponding to the target object and a second index set not corresponding to the target object, A medical image processing device characterized in that a first weight corresponding to the first index set and a second weight corresponding to the second index set are determined as weights for the target object.

7. In paragraph 5, The above processor, A medical image processing device characterized in that the regions of the target object included in the first 3D image are classified into the plurality of classes through K-mean clustering.

8. In paragraph 7, The above processor, Extract the index values ​​with the largest pixel value in a given direction, A medical image processing device characterized in that the class map is generated based on the extracted index values ​​and pixel values ​​corresponding to each of the extracted index values.

9. In paragraph 5, The above multiple classes are, A first class corresponding to the first part located on the X-ray generator side; A second class corresponding to a second part located on the X-ray detector side; and A medical image processing device characterized by including a third class corresponding to an uncertainty portion.

10. In paragraph 6, The above processor, For each of the above multiple classes, the average and standard deviation of the index values ​​for the given direction are calculated, A medical image processing device characterized in that the plurality of index values ​​are classified into the first index set and the second index set based on the average and the standard deviation calculated for each of the plurality of classes.

11. In paragraph 2, The above plurality of projection data are generated based on detection of the X-rays irradiated from a plurality of X-ray sources, The above processor, Apply log projection to each of the above multiple result data, Apply weighted projection to the above log-projected result data, Apply a bidirectional ramp filter to the above weighted projected result data, A medical image processing device characterized in that it generates the first three-dimensional image reconstructed based on each of the result data to which the bidirectional lamp filter is applied.

12. In paragraph 6, The above processor, From the above second 3D image, a plurality of image pieces sliced ​​along a predetermined direction are obtained, A medical image processing device characterized in that it generates the two-dimensional image by applying one of the first weight and the second weight to each of the plurality of image fragments.

13. In paragraph 12, The above processor, A medical image processing device characterized in that it generates the two-dimensional image by performing weighted summation by applying the first weight and any one of the above to each of the plurality of image fragments.

14. In paragraph 13, The above processor, Based on the following mathematical expression 1, the first weight and the second weight are calculated, [Mathematical Formula 1] A medical image processing device, characterized in that the E is a weighted mask according to the specific gravity of the target object, the A is a first index set, the W is a mask corresponding to the first 3D image, λ is a trade-off parameter, and I is an identity matrix.

15. In paragraph 14, The above processor, Based on the following mathematical expressions 2 and 3, the weighted summation is performed, [Equation 2] [Equation 3] A medical image processing device, wherein the above f is the second 3D image and the above S is the 2D image.

Images