Image processing device, method, and program

Abstract

A processor performs processes of: acquiring two radiation images acquired by photographing a subject including a bone and a soft tissue with radiation having different energy distributions; deriving a bone image including a microstructure inside the bone on the basis of the two radiation images; and deriving a likelihood of osteoporosis based on the bone image or a partial image of the bone image including the target bone.

Description

Technical Field

[0001] The present invention relates to an image processing device, method and program. Background Art

[0002] As a complication of osteoporosis, fractures of the femur and vertebrae are likely to leave patients bedridden. The five-year survival rate for bedridden patients is known to be lower than that of cancer. With the global prevalence of osteoporosis, early detection and treatment of osteoporosis are expected to reduce medical and nursing costs. Therefore, various methods for assessing osteoporosis have been proposed.

[0003] For example, the specification of Chinese Patent Gazette No. 113129286 proposes a method for detecting osteoporosis by obtaining a bone density image of uniform size from a bone density image obtained by a quantitative CT method using DXA (Dual X-ray Absorptiometry) or CT (Computed Tomography) images, and performing feature recognition on the bone density image of uniform size. Summary of the Invention

[0004] Technical issues to be solved by the invention

[0005] On the other hand, bones contain trabecular structures of cancellous bone within them. Trabeculae are fine structures. While healthy individuals have densely arranged trabeculae, the density of trabeculae in osteoporosis patients is reduced. Therefore, accurately understanding the density of the trabecular microstructure helps to detect osteoporosis with high precision. However, the spatial resolution of bone density images obtained by the aforementioned DXA method or quantitative CT method is large. For example, the spatial resolution of bone density images obtained by the DXA method is approximately 0.6 to 2.0 mm / pixel, while the spatial resolution of bone density images obtained by the quantitative CT method is approximately 0.5 to 1.0 mm / pixel. In bone density images obtained by the DXA method or quantitative CT method with such spatial resolution, the fine structure of the trabeculae is not fully depicted within the image, making it impossible to accurately determine osteoporosis.

[0006] The present invention has been made in view of the above-mentioned circumstances, and an object of the present invention is to enable simple and highly accurate determination of osteoporosis.

[0007] Means for solving technical problems

[0008] The image processing device according to the present invention comprises at least one processor.

[0009] The processor performs the following processing:

[0010] acquiring two radiographic images obtained by imaging a subject with radiation, wherein the radiation has different energy distributions and the subject includes bones and soft tissue;

[0011] deriving a bone image including the fine structure inside the bone based on the two radiographic images; and

[0012] The possibility of osteoporosis is derived based on the bone image or a partial image of the bone image including the target bone.

[0013] The “partial image including the target bone” may be an image of only the target bone, or an image including the target bone and its surrounding area.

[0014] In addition, in the image processing apparatus according to the present invention, the bone image may include a trabecular structure of cancellous bone included in the bone.

[0015] Furthermore, in the image processing apparatus according to the present invention, the trabecular structure of cancellous bone can be depicted in the skeletal image.

[0016] "Delineating" means visually identifying the trabecular structure of cancellous bone in the skeletal image. To depict the trabecular structure of cancellous bone, the spatial resolution per pixel of the skeletal image must be less than or equal to the spatial resolution of the trabecular structure of cancellous bone. For example, the spatial resolution of the skeletal image is 0.2 mm / pixel or less, preferably 0.15 mm / pixel or less, and more preferably 0.1 mm / pixel or less.

[0017] Furthermore, in the image processing apparatus according to the present invention, the processor may extract a partial image from at least one of the two radiographic images and the bone image.

[0018] Furthermore, in the image processing apparatus according to the present invention, the processor may extract at least one target bone from at least one of the two radiographic images and the bone image, and extract the partial image based on the center of gravity of the target bone.

[0019] Furthermore, in the image processing apparatus according to the present invention, the processor may function as a trained model that has been subjected to machine learning so as to output the possibility of osteoporosis by inputting a bone image or a partial image.

[0020] Furthermore, in the image processing device according to the present invention, the processor can function as a trained model that has undergone machine learning so as to output the possibility of osteoporosis by inputting the photographic conditions of the two radiographic images in addition to the bone image or the partial image.

[0021] Furthermore, in the image processing apparatus according to the present invention, the processor may deduce the possibility of osteoporosis based on statistical values ​​of pixel values ​​of the target bone included in the bone image or the partial image.

[0022] Furthermore, in the image processing apparatus according to the present invention, the statistical value may be at least one of an average value, a standard deviation, a maximum value, and a minimum value of pixel values ​​in the target bone and each surrounding area of ​​the target bone.

[0023] Furthermore, in the image processing device according to the present invention, the processor can derive the possibility by performing weighted operations on multiple statistical values ​​including the average value, standard deviation, maximum value and minimum value of pixel values ​​in the target bone and each surrounding area of ​​the target bone.

[0024] Furthermore, in the image processing apparatus according to the present invention, the weights of the weighted operation can be derived through machine learning.

[0025] Furthermore, in the image processing apparatus according to the present invention, the processor may also notify the possibility.

[0026] Furthermore, in the image processing apparatus according to the present invention, when the probability is equal to or greater than a preset threshold, the processor may issue a warning.

[0027] Furthermore, in the image processing apparatus according to the present invention, the processor may determine the priority of treatment of osteoporosis based on the probability and notify the priority.

[0028] Furthermore, in the image processing apparatus according to the present invention, the processor may remove scattered ray components included in the two radiographic images, thereby deriving the bone image.

[0029] Furthermore, in the image processing apparatus according to the present invention, the processor may perform the following processing:

[0030] in a soft tissue region including only soft tissue in the two radiographic images, deriving a characteristic of the soft tissue related to attenuation of radiation based on the two radiographic images;

[0031] deriving characteristics of soft tissue in a bone region including bones in two radiographic images based on characteristics of soft tissue derived in a soft tissue region surrounding the bone region; and

[0032] An image in which bones are emphasized is derived as a bone image based on characteristics of at least soft tissue in the region of the subject in the two radiographic images.

[0033] Also, in the image processing apparatus according to the present invention, the target bone may be a femur or a vertebra.

[0034] In the image processing method according to the present invention, the computer performs the following processing:

[0035] acquiring two radiographic images obtained by imaging a subject with radiation, wherein the radiation has different energy distributions and the subject includes bones and soft tissue;

[0036] deriving a bone image including the fine structure inside the bone based on the two radiographic images; and

[0037] The possibility of osteoporosis is derived based on the bone image or a partial image of the bone image including the target bone.

[0038] The image processing program according to the present invention causes a computer to execute the following steps:

[0039] acquiring two radiographic images obtained by imaging a subject with radiation, wherein the radiation has different energy distributions and the subject includes bones and soft tissue;

[0040] deriving a bone image including the fine structure inside the bone based on the two radiographic images; and

[0041] The possibility of osteoporosis is derived based on the bone image or a partial image of the bone image including the target bone.

[0042] Effects of the Invention

[0043] According to the present invention, osteoporosis can be determined simply and accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a schematic block diagram showing the configuration of a radiographic imaging system to which an image processing apparatus according to a first embodiment of the present invention is applied.

[0045] Figure 2 This is a diagram showing a schematic configuration of an image processing device according to the first embodiment.

[0046] Figure 3 This is a diagram showing the functional configuration of the image processing device according to the first embodiment.

[0047] Figure 4 is a diagram showing a skeleton image.

[0048] Figure 5 is a diagram showing the segmentation result.

[0049] Figure 6 : is a diagram showing a schematic structure of a neural network used in this embodiment.

[0050] Figure 7This is a diagram showing training data used to construct a trained model in the first embodiment.

[0051] Figure 8 This is a diagram showing an example of a trained model used in the first embodiment.

[0052] Figure 9 It is a diagram showing an example of a display screen.

[0053] Figure 10 This is a flowchart of the learning process performed in the first embodiment.

[0054] Figure 11 Graph showing the distribution of radiographic images acquired in this embodiment.

[0055] Figure 12 : is a graph showing the distribution of bone images derived from two radiographic images acquired in this embodiment.

[0056] Figure 13 This is a graph showing the distribution of bone images (equivalent to bone density images) acquired by the quantitative CT method.

[0057] Figure 14 Graph showing the distribution of bone images (equivalent to bone density images) acquired by the DXA method.

[0058] Figure 15 Is used to illustrate Figures 11 to 14 The horizontal axis of the graph shown is a graph of the intervals represented.

[0059] Figure 16 This is a diagram showing the functional configuration of an image processing device according to the second embodiment.

[0060] Figure 17 This is a diagram showing training data used to construct a trained model in the second embodiment.

[0061] Figure 18 This is a diagram showing an example of a trained model used in the second embodiment.

[0062] Figure 19 This is a diagram showing the functional configuration of an image processing device according to a third embodiment.

[0063] Figure 20 It is a diagram showing an example of a display screen including a warning.

[0064] Figure 21 2 is a diagram showing an example of a display screen including treatment priorities.

[0065] Figure 22This is a diagram showing the functional structure of a skeleton image derivation unit of an image processing device according to a fourth embodiment.

[0066] Figure 23 FIG. 1 is a diagram schematically showing processing performed in the skeleton image derivation unit of the image processing device according to the fourth embodiment.

[0067] Figure 24 This is a flowchart showing the processing performed in the fourth embodiment. DETAILED DESCRIPTION

[0068] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. Figure 1 1 is a schematic block diagram showing the configuration of a radiographic imaging system to which an image processing apparatus according to a first embodiment of the present invention is applied. Figure 1 As shown, the radiation imaging system according to the present embodiment includes an imaging device 1 and an image processing device 10 according to the present embodiment.

[0069] The imaging device 1 is an imaging device that performs energy subtraction using the so-called one-shot method. The imaging device irradiates the radiation such as X-rays emitted from the radiation source 3 and transmitted through the subject H with different energies onto the first radiation detector 5 and the second radiation detector 6. Figure 1 As shown, a first radiation detector 5, a radiation energy conversion filter 7 made of a copper plate or the like, and a second radiation detector 6 are arranged in this order from the side closest to the radiation source 3 to drive the radiation source 3. The first radiation detector 5 and the second radiation detector 6 are in close contact with the radiation energy conversion filter 7.

[0070] Thus, the first radiation detector 5 acquires a first radiographic image G1 of the subject H using low-energy radiation that also includes so-called soft rays. Furthermore, the second radiation detector 6 acquires a second radiographic image G2 of the subject H using high-energy radiation from which the soft rays have been removed. Furthermore, both the first radiographic image G1 and the second radiographic image G2 are two-dimensional images, being transmission images of the subject acquired through simple radiographic imaging by irradiating the subject H with radiation once. Therefore, both the first radiographic image G1 and the second radiographic image G2 are simple radiographic images. Furthermore, in the first radiographic image G1 and the second radiographic image G2, the higher the brightness (i.e., the lower the density), the larger the pixel value.

[0071] The first radiation detector 5 and the second radiation detector 6 are capable of repeatedly recording and reading out radiation images. They may be direct radiation detectors, which generate charge by directly receiving radiation, or indirect radiation detectors, which first convert radiation into visible light and then convert the visible light into a charge signal. The radiation image signal readout method preferably uses a so-called TFT readout method, which reads out the radiation image signal by turning on and off a TFT (thin film transistor) switch, or a so-called optical readout method, which reads out the radiation image signal by irradiating a reading light beam. However, the method is not limited to these methods, and other methods may also be used.

[0072] Furthermore, the radiation image processing apparatus 10 is connected to the image storage system 9 via a network (not shown).

[0073] The image storage system 9 stores the image data of radiographic images captured by the imaging device 1. From the stored radiographic images, the image storage system 9 extracts images corresponding to requests from the image processing device 10 and transmits them to the requesting device. A specific example of the image storage system 9 is a PACS (Picture Archiving and Communication Systems).

[0074] Next, the image processing device according to the first embodiment will be described. Figure 2 The hardware structure of the image processing device involved in this embodiment is described. Figure 2 As shown, image processing device 10 is a computer such as a workstation, server, or personal computer, and includes a CPU (Central Processing Unit) 11, a nonvolatile storage device 13, and a memory 16 serving as a temporary storage area. Furthermore, image processing device 10 includes a display 14 such as a liquid crystal display, input devices 15 such as a keyboard and a mouse, and a network interface (I / F) 17 for connecting to a network (not shown). CPU 11, storage device 13, display 14, input device 15, memory 16, and network I / F 17 are connected to a bus 18. CPU 11 is an example of a processor in the present invention.

[0075] The storage device 13 is implemented by an HDD (Hard Disk Drive), an SSD (Solid State Drive), a flash memory, or the like. The storage device 13, serving as a storage medium, stores the image processing program 12 installed in the image processing apparatus 10. The CPU 11 reads the image processing program 12 from the storage device 13, expands it into the memory 16, and executes the expanded image processing program 12.

[0076] The image processing program 12 is stored in a storage device of a server computer connected to a network or in a network storage device in a state that allows external access, and is downloaded upon request and installed in the computer constituting the image processing apparatus 10. Alternatively, the image processing program 12 may be distributed by being recorded on a recording medium such as a DVD (Digital Versatile Disc) or a CD-ROM (Compact Disc Read Only Memory), and then installed from the recording medium into the computer constituting the image processing apparatus 10.

[0077] Next, the functional configuration of the image processing device according to the first embodiment will be described. Figure 3 1 is a diagram showing the functional structure of the image processing device according to the first embodiment. Figure 3 As shown, the image processing device 10 includes an image acquisition unit 21, a scattered ray removal unit 22, a bone image derivation unit 23, a segmentation unit 24, a likelihood derivation unit 25, and a display control unit 26. Furthermore, the CPU 11, by executing the image processing program 12, functions as the image acquisition unit 21, the scattered ray removal unit 22, the bone image derivation unit 23, the segmentation unit 24, the likelihood derivation unit 25, and the display control unit 26, and also functions as a trained model 25A, which will be described later.

[0078] The image acquisition unit 21 causes the imaging device 1 to image the subject H, thereby acquiring, for example, a first radiographic image G1 and a second radiographic image G2, which are frontal images of the vicinity of the crotch of the subject H, from the first and second radiation detectors 5 and 6. When acquiring the first and second radiographic images G1 and G2, imaging conditions are set, including the imaging dose, tube voltage, the SID (Source Image Receptor Distance) (the distance between the radiation source 3 and the surfaces of the first and second radiation detectors 5 and 6), the SOD (Source Object Distance) (the distance between the radiation source 3 and the surface of the subject H), and the presence or absence of a scattered radiation removal grid.

[0079] As described later, SOD and SID are used to calculate body thickness distribution. SOD is preferably acquired using, for example, a TOF (Time of Flight) camera, while SID is preferably acquired using, for example, a potentiometer, ultrasonic rangefinder, or laser rangefinder.

[0080] The imaging conditions can be set by an operator inputting them through the input device 15. The set imaging conditions are stored in the storage device 13.

[0081] In the present embodiment, the first radiographic image G1 and the second radiographic image G2 may be acquired by a program different from the image processing program 12 and stored in the storage device 13. In this case, the image acquisition unit 21 acquires these images for processing by reading the first radiographic image G1 and the second radiographic image G2 stored in the storage device 13 from the storage device 13.

[0082] Here, there is cancellous bone in the human skeleton, and this cancellous bone is located inside the cortical bone covering the outside of the skeleton, and has a dense structure. Cancellous bone has a structure (hereinafter referred to as trabecular structure) formed by three-dimensional connection of trabeculae with a thickness of about 150 μm. The first radiographic image G1 and the second radiographic image G2 obtained in the present embodiment include the trabecular structure of the cancellous bone contained in the interior of the skeleton. That is, the first radiographic image G1 and the second radiographic image G2 are high-definition images, in which the trabecular structure of the cancellous bone is depicted so that the trabecular structure of the cancellous bone can be visually identified. In order to be able to depict the trabecular structure of this cancellous bone, the spatial resolution of the first radiographic image G1 and the second radiographic image G2 is less than 0.2 mm / pixel, preferably 0.15 mm / pixel, and more preferably less than 0.1 mm / pixel.

[0083] Here, each of the first and second radiographic images G1 and G2 includes, in addition to primary radiation components of radiation that has passed through the subject H, scattered radiation components based on radiation scattered within the subject H. Therefore, the scattered radiation removal unit 22 removes the scattered radiation components from the first and second radiographic images G1 and G2. For example, the scattered radiation removal unit 22 can apply the method described in Japanese Patent Application Laid-Open No. 2015-043959 to remove the scattered radiation components from the first and second radiographic images G1 and G2. When using the method described in Japanese Patent Application Laid-Open No. 2015-043959, the body thickness distribution of the subject H is deduced simultaneously with the deduction of the scattered radiation components for removing the scattered radiation components.

[0084] The following describes a method for removing scattered radiation components from the first radiographic image G1. However, the same method can be used to remove scattered radiation components from the second radiographic image G2. First, the scattered radiation removal unit 22 acquires a virtual model K of the subject H having an initial thickness distribution T0(x, y). The virtual model K is data that virtually represents the subject H, with the thickness of the initial thickness distribution T0(x, y) associated with the coordinate position of each pixel in the first radiographic image G1. The virtual model K of the subject H having the initial thickness distribution T0(x, y) can be pre-stored in the storage device 13. Alternatively, the thickness distribution T(x, y) of the subject H can be calculated based on the SID and SOD included in the imaging conditions. In this case, the thickness distribution can be determined by subtracting the SOD from the SID.

[0085] Next, the scattered ray removing unit 22 generates an image obtained by synthesizing an estimated primary radiographic image obtained by imaging the virtual model K and an estimated scattered ray image based on the virtual model K, as an estimated image estimated from the first radiographic image G1 obtained by imaging the subject H. The estimated primary radiographic image is an image estimated from the primary radiographic image obtained by imaging the virtual model K, and the estimated scattered ray image is an image estimated from the scattered ray image obtained by imaging the virtual model K.

[0086] Next, the scattered ray removal unit 22 corrects the initial thickness distribution T0(x, y) of the virtual model K to reduce the difference between the estimated image and the first radiographic image G1. The scattered ray removal unit 22 repeatedly generates the estimated image and corrects the thickness distribution until the difference between the estimated image and the first radiographic image G1 satisfies a predetermined termination condition. The scattered ray removal unit 22 derives the thickness distribution when the termination condition is satisfied as the thickness distribution T(x, y) of the subject H. Furthermore, the scattered ray removal unit 22 removes the scattered ray components contained in the first radiographic image G1 by subtracting the scattered ray components when the termination condition is satisfied from the first radiographic image G1. In the following description, it is assumed that the scattered ray components of the first radiographic image G1 and the second radiographic image G2 have been removed.

[0087] The skeleton image derivation unit 23 performs energy subtraction processing to derive a skeleton image Gb from the first radiographic image G1 and the second radiographic image G2, in which the skeleton of the subject H has been extracted. When deriving the skeleton image Gb, the skeleton image derivation unit 23 performs weighted subtraction operation between corresponding pixels on the first radiographic image G1 and the second radiographic image G2 as shown in the following formula (1). Thus, Figure 4 As shown in FIG, the skeleton image Gb of the skeleton of the subject H included in each of the radiographic images G1 and G2 is extracted. In formula (1), β1 is a weighting coefficient. In addition, Figure 4A partial image B described later is shown in FIG.

[0088] Gb(x, y)=G1(x, y)-β1×G2(x, y) (1)

[0089] The segmentation unit 24 segments the skeletal image Gb into a femur region, a pelvis region, and a vertebrae region. Segmentation can be performed using an extraction model that has undergone machine learning to extract the femur, pelvis, and vertebrae from the skeletal image Gb. Alternatively, templates representing the femur, pelvis, and vertebrae, respectively, may be pre-stored in the storage device 13, and segmentation can be performed by performing template matching between these templates and the skeletal image Gb.

[0090] Figure 5 2 is a diagram showing the result of segmentation performed by the segmentation unit 24. Figure 5 As shown in FIG. 1 , the skeleton region in the skeleton image Gb is divided into a femur region A1, a pelvis region A2, and a vertebra region A3. Figure 5 In FIG, the segmentation result is shown by giving different hatching to the femur region A1, the pelvis region A2, and the vertebra region A3.

[0091] On the other hand, regarding the vertebrae, the skeleton image Gb only includes the sacrum and lumbar vertebrae. Anatomically, the lumbar vertebrae are divided into the fifth lumbar vertebra L5, the fourth lumbar vertebra L4, the third lumbar vertebra L3, the second lumbar vertebra L2, and the first lumbar vertebra L1, from the pelvic side toward the neck. Therefore, the segmentation unit 24 preferably segments the sacrum and the five lumbar vertebrae into distinct regions.

[0092] Alternatively, the segmentation unit 24 may segment only the target bone in the skeletal image Gb. For example, in this embodiment, since the femur is the target bone, only the femoral region A1 may be segmented. Furthermore, the segmentation unit 24 may segment the first radiographic image G1 or the second radiographic image G2 instead of the skeletal image Gb.

[0093] The possibility derivation unit 25 derivates the possibility of osteoporosis based on the bone image Gb or the partial image including the target bone in the bone image Gb. When derivation of the possibility based on the partial image, the possibility derivation unit 25 extracts the partial image from the bone image Gb. For example, Figure 4 As shown, in the skeleton image Gb, the partial image B includes the femur, the target bone extracted by the segmentation unit 24, and its surrounding area. Alternatively, the partial image may include only the femur region, the target bone. Furthermore, the likelihood derivation unit 25 may derive the center of gravity of the femur and extract a predetermined range with the center of gravity of the femur as the partial image.

[0094] In addition, the target bone may be a vertebra. At this time, the possibility derivation unit 25 deduces the possibility of osteoporosis based on the partial image including the vertebra. Therefore, the possibility derivation unit 25 extracts the partial image including the vertebra as the target bone. At this time, the center of gravity of the vertebra can be derived, and the partial image can be derived so that the center of gravity is used as the center of gravity of the partial image. In addition, the center of gravity position of each of the multiple vertebrae (for example, the second lumbar vertebra, the third lumbar vertebra, and the fourth lumbar vertebra) can be derived, and the partial image can be derived so that the center of gravity position of the derived center of gravity position becomes the center of gravity of the partial image.

[0095] In this embodiment, the likelihood derivation unit 25 derivates the likelihood of osteoporosis using the trained model 25A. In the trained model 25A, when the partial image B is input, the likelihood P0 of osteoporosis is output.

[0096] Trained model 25A is constructed by performing machine learning on a neural network using a partial image of the target bone within skeletal image Gb and accurate data related to skeletal image Gb indicating the presence or absence of osteoporosis as training data. Osteoporosis reduces the density of the trabecular structure of the cancellous bone within the skeleton. Therefore, the accurate training data is determined by a doctor by reading skeletal image Gb and assessing the condition of the trabecular structure, thereby determining whether osteoporosis is present.

[0097] Examples of neural networks include simple perceptrons, multilayer perceptrons, deep neural networks, convolutional neural networks, deep belief networks, recursive neural networks, and probabilistic neural networks. In this embodiment, a convolutional neural network is used as the neural network.

[0098] Figure 6 is a diagram showing a neural network used in this embodiment. Figure 6 As shown, the neural network 60 includes an input layer 61, an intermediate layer 62, and an output layer 63. The intermediate layer 62 includes, for example, a plurality of convolutional layers 65, a plurality of pooling layers 66, and a fully connected layer 67. In the neural network 60, the fully connected layer 67 is located before the output layer 63. Furthermore, in the neural network 60, the convolutional layers 65 and the pooling layers 66 are alternately arranged between the input layer 61 and the fully connected layer 67.

[0099] In addition, the structure of the neural network 60 is not limited to Figure 6 For example, the neural network 60 may include one convolutional layer 65 and one pooling layer 66 between the input layer 61 and the fully connected layer 67 .

[0100] Figure 7 : is a diagram showing an example of training data used for learning a neural network in the first embodiment. Figure 7As shown in FIG. 4 , the training data 40 is composed of learning data 41 and correct data 42. The learning data 41 is composed of a partial image 43 including the target bone in the bone image. Figure 7 , the partial image includes the femur as the target bone and its surrounding area. The partial image 43 is derived from the bone image of a healthy person and the bone image of a patient with osteoporosis.

[0101] The presence or absence of osteoporosis in the correct data 42 can be set according to whether the partial image 43 is an image derived from a bone image of a healthy person or an image derived from a bone image of an osteoporosis patient.

[0102] Correct data 42 is composed of the presence or absence of osteoporosis. In this embodiment, the probability P0 takes a value between 0 and 1. Therefore, the correct data 42 value is 1 if the patient has osteoporosis, and the correct data 42 value is 0 if the patient does not have osteoporosis.

[0103] Furthermore, the partial image 43 serving as the learning data 41 can be derived by processing a bone image of a healthy person according to the presence or absence of osteoporosis. This increases the amount of training data and enables efficient learning.

[0104] If the pixel value B(x, y) of each pixel of the partial image B including the target bone in the bone image Gb of the patient H is input into the trained model 25A constructed by machine learning of the neural network using a large amount of such training data 40, then Figure 8 As shown, the trained model 25A outputs the possibility P0 that the subject H suffers from osteoporosis.

[0105] Furthermore, when the likelihood derivation unit 25 deduces the likelihood P0 of osteoporosis based on the bone image Gb instead of the partial image, the trained model 25A is constructed so that, when the bone image Gb is input, the likelihood P0 of osteoporosis is output. The learning data 41 of the training data 40 used in this case is the bone images of healthy individuals and osteoporosis patients.

[0106] In order to notify the possibility P0 of osteoporosis derived by the possibility deriving unit 25 , the display control unit 26 displays the possibility P0 on the display 14 . Figure 9 FIG is a diagram showing a display screen indicating the possibility of osteoporosis. Figure 9 As shown, the bone image Gb of the subject H and the probability 51 of osteoporosis are displayed on the display screen 50. In addition, the probability takes a value of 0 to 1, but Figure 9 In the display screen 50 shown, the probability is expressed as a percentage.

[0107] Next, the processing performed in the first embodiment will be described. Figure 10 This is a flowchart illustrating the processing performed in the first embodiment. Furthermore, a first radiographic image G1 and a second radiographic image G2 are acquired through imaging and stored in the storage device 13. When an instruction to start processing is input from the input device 15, the image acquisition unit 21 acquires the first radiographic image G1 and the second radiographic image G2 from the storage device 13 (radiation image acquisition; step ST1).

[0108] Next, the scattered radiation removal unit 22 removes scattered radiation components from the first and second radiographic images G1 and G2 (step ST2). Furthermore, the bone image derivation unit 23 derivates a bone image Gb that includes the microstructure of the bone interior of the subject H from the first and second radiographic images G1 and G2 (step ST3). Furthermore, the segmentation unit 24 segments the target bone, the femur, within the bone image Gb (step ST4).

[0109] Next, the likelihood derivation unit 25 extracts the partial image B including the target bone from the skeleton image Gb and deduces the likelihood P0 of osteoporosis based on the partial image B (step ST5). The display control unit 26 then displays the likelihood P0 of osteoporosis derived by the likelihood derivation unit 25 on the display 14 (step ST6), and the process ends.

[0110] Figures 11 to 14 This is a diagram for explaining the spatial resolution corresponding to the type of image. Figure 11 is a graph showing the distribution of radiographic images acquired in this embodiment, Figure 12 is a graph showing the distribution of bone images derived from two radiographic images acquired in this embodiment, Figure 13 is a graph showing the distribution of bone images (equivalent to bone density images) acquired by quantitative CT. Figure 14 This is a graph showing the distribution of bone images (equivalent to bone density images) acquired by the DXA method.

[0111] In addition, if Figure 15 As shown, Figures 11 to 14 The horizontal axis of the graph shown represents the position within interval 70 of the femoral neck, and the vertical axis represents the pixel value of interval 70. The vertical lines are spaced 10 mm apart. Furthermore, the spatial resolution of the radiographic image and the bone image is 0.15 mm / pixel, which is equivalent to the spatial resolution of 1.0 mm / pixel for bone density images acquired using quantitative CT, and 2.0 mm / pixel for bone density images acquired using DXA.

[0112] like Figure 11As shown in the graph, in the radiographic image acquired in this embodiment, the pixel values ​​vary slightly within the bone in interval 70, and thus it can be seen that the trabecular structure of the cancellous bone is depicted. Figure 12 As shown in the graph, in the bone image obtained in this embodiment, the pixel value changes slightly inside the bone in interval 70, and Figure 11 Therefore, it can be seen that the trabecular structure of the cancellous bone is more clearly depicted in the skeletal image acquired in this embodiment than in the radiographic image.

[0113] On the other hand, Figure 13 As shown in the graph, in the bone image equivalent to the bone density image obtained by the quantitative CT method, there is almost no slight change in pixel value inside the bone in interval 70, so it can be seen that the trabecular structure of the cancellous bone is hardly depicted. Figure 14 As shown in the graph, in the bone image equivalent to the bone density image obtained by the DXA method, no slight change in pixel value is observed inside the bone in interval 70, so it can be seen that the trabecular structure of the cancellous bone is not depicted at all.

[0114] In the first embodiment, a bone image Gb including the microstructure inside the bone is derived, and the possibility P0 of osteoporosis is derived based on a partial image B including the target bone in the bone image Gb. Figure 12 As shown in the graph, the following method is used to deduce the probability P0 of osteoporosis from a skeletal image Gb that clearly depicts the fine trabecular structure of the cancellous bone within the skeleton. Here, the first radiographic image G1 and the second radiographic image G2 used to obtain the skeletal image Gb are acquired through simple radiographic imaging. Therefore, according to this embodiment, osteoporosis can be determined simply and accurately.

[0115] Furthermore, since scattered radiation components are removed from the first and second radiation images G1 and G2, a bone image Gb with less image blurring due to scattered radiation can be derived. Therefore, the likelihood P0 of osteoporosis can be derived with higher accuracy.

[0116] In this case, the scattered ray removing unit 22 is unnecessary.

[0117] Next, a second embodiment of the present invention will be described. Figure 162 is a diagram illustrating the functional configuration of an image processing device according to a second embodiment. In the first embodiment, the likelihood derivation unit 25 uses a trained model 25A that has undergone machine learning so that, when a skeletal image Gb or a partial image of a skeletal image Gb including a target bone is input, the likelihood of osteoporosis P0 is output. The image processing device 10A according to the second embodiment differs from the first embodiment in that, in addition to the skeletal image Gb or the partial image B of the skeletal image Gb including a target bone, it also uses a trained model 25B that has undergone machine learning so that, when the imaging conditions used to acquire the first radiographic image G1 and the second radiographic image G2 are input, the likelihood of osteoporosis P0 is output.

[0118] Figure 17 : is a diagram showing an example of training data used for learning a neural network in the second embodiment. Figure 17 As shown, in the second embodiment, the training data 40A used for learning the neural network includes Figure 7 In addition to the training data 40 shown, the imaging conditions 44 when acquiring the learning data 41 are also included.

[0119] If the pixel value B(x, y) of each pixel of the bone image Gb of the patient's subject H or the partial image of the bone image Gb including the target bone, and the imaging condition S0 when the first radiographic image G1 and the second radiographic image G2 used to generate the bone image Gb are input into the trained model 25B constructed by machine learning of the neural network using a large amount of such training data 40A, then Figure 18 As shown, the trained model 25B outputs the possibility P0 that the subject H suffers from osteoporosis.

[0120] Next, a third embodiment of the present invention will be described. Figure 19 This figure illustrates the functional configuration of an image processing device according to a third embodiment. In the first and second embodiments, the likelihood derivation unit 25 deduces the likelihood P0 of osteoporosis using trained models 25A and 25B. The image processing device 10B according to the third embodiment differs from the first and second embodiments in that the likelihood derivation unit 25 deduces the likelihood P0 of osteoporosis based on statistical values ​​of a portion of the image including the target bone in the skeleton image Gb.

[0121] In the third embodiment, the likelihood derivation unit 25 derives statistical values ​​for the skeleton image Gb or the partial image B of the skeleton image Gb including the target bone. For example, the likelihood derivation unit 25 derives as a statistical value at least one of the following values: the average value μA, variance value σA, maximum value MaxA, and minimum value MinA of the pixel values ​​of the region containing only the target bone, as well as the average value μB, variance value σB, maximum value MaxB, and minimum value MinB of the region surrounding only the target bone, in the partial image B including the region containing the target bone and its surrounding area.

[0122] Here, when osteoporosis develops, the density of the trabecular structure of the cancellous bone decreases, resulting in a decrease in the number of high-brightness (low-concentration) components, and thus a decrease in the average values ​​μA and μB. Furthermore, a decrease in the density of the trabecular structure of the cancellous bone increases the difference in density between the trabeculae, i.e., the difference in pixel values, and thus a decrease in the variance values ​​σA and σB. Furthermore, a decrease in the density of the trabecular structure of the cancellous bone decreases the number of high-brightness (low-concentration) components, and thus a decrease in the maximum values ​​MaxA and MaxB and the minimum values ​​MinA and MinB.

[0123] The likelihood derivation unit 25 can derive the likelihood P0 of osteoporosis by normalizing any of these eight statistical values ​​to 0 to 1. However, in the third embodiment, the likelihood P0 is derived by weighting these eight statistical values ​​using the following equation (2). In equation (2), C1 to C8 are weight coefficients for each of the eight statistical values. Furthermore, among the weight coefficients C1 to C8, the weight coefficients C1 and C2 for the average μA and variance σA of the pixel values ​​in the target bone region alone are set to values ​​greater than the other weight coefficients C3 to C8.

[0124] [Formula 1]

[0125]

[0126] In addition, the weight coefficients C1 to C8 in formula (2) can be obtained by performing machine learning on a classifier such as the AdaBoost algorithm or SVM. Alternatively, a discriminant analysis method using the Mahalanobis distance can be used. For example, when using the AdaBoost algorithm, the above-mentioned statistical values ​​are derived for a large number of osteoporosis images and a large number of normal images, and the weight coefficients multiplied by each statistical value are optimized so that the discrimination accuracy of osteoporosis is maximized for the derived statistical value (i.e., the error rate is minimized).

[0127] In the third embodiment, Figure 12As shown in the graph, the fine trabecular structure of the spongy bone within the skeleton is clearly depicted, and the probability P0 of osteoporosis is derived from the bone image Gb derived from the first radiographic image G1 and the second radiographic image G2 acquired by simple imaging. Therefore, similar to the first embodiment, according to the third embodiment, osteoporosis can be determined simply and accurately.

[0128] In addition, in each of the above embodiments, Figure 9 As shown, a display screen 50 including the bone image Gb and the probability 51 of osteoporosis is displayed on the display 14, but the present invention is not limited to this. If the probability P0 derived by the probability derivation unit 25 exceeds a predetermined threshold, it indicates that osteoporosis is progressing, and a warning can be issued. The comparison of the probability P0 with the threshold can be performed by the probability derivation unit 25, the display control unit 26, or a dedicated comparison function can be provided in the image processing device 10.

[0129] Figure 20 FIG is a diagram showing a display screen including a warning. Figure 20 As shown, the display screen 50A includes a warning display 52 in addition to the skeleton image Gb and the possibility of osteoporosis 51. The warning display 52 includes a text "Possibility of osteoporosis is high".

[0130] In this manner, by issuing this warning when the probability P0 is equal to or greater than a preset threshold value, a doctor can easily recognize that the patient from whom the skeleton image Gb was derived has a high probability of suffering from osteoporosis.

[0131] Furthermore, the treatment priority for osteoporosis may be determined based on the probability P0 of osteoporosis derived by the probability derivation unit 25, and the treatment priority may be displayed in addition to the bone image Gb and the probability 51 of osteoporosis. The determination of the treatment priority may be performed by the probability derivation unit 25 or by the display control unit 26, or a dedicated function for determining the treatment priority may be provided in the image processing device 10.

[0132] Here, in this embodiment, the probability P0 of osteoporosis is a value between 0 and 1. Therefore, when the probability P0 is between 0.7 and 1.0, the priority is determined to be "high", when the probability P0 is between 0.3 and 0.7, the priority is determined to be "medium", and when the probability P0 is between 0.0 and 0.3, the priority is determined to be "low".

[0133] Figure 21 FIG is a diagram showing a display screen including treatment priorities. Figure 21As shown, the display screen 50B includes a treatment priority 53 in addition to the bone image Gb and the possibility 51 of osteoporosis. Figure 21 Medium, treatment priority 53 is "High".

[0134] In this way, the priority of treatment is determined based on the probability P0 of osteoporosis, and the determined priority is notified, thereby being used for triage of patients.

[0135] Furthermore, in each of the above-described embodiments, the skeletal image derivation unit 23 derives the skeletal image Gb by performing a weighted subtraction operation on the first radiographic image G1 and the second radiographic image G2 using equation (1). However, the present invention is not limited to this embodiment. Alternatively, the skeletal image Gb may be derived using characteristics related to the attenuation of radiation in the bones and soft tissues of the subject H. This will be described below as a fourth embodiment.

[0136] Figure 22 : is a diagram showing the functional structure of the skeleton image derivation unit in the fourth embodiment. Figure 22 As shown, in the fourth embodiment, the skeleton image derivation unit 23 includes a region identification unit 31 , a characteristic derivation unit 32 , and an image derivation unit 33 .

[0137] Here, the first radiographic image G1 and the second radiographic image G2 acquired by the image acquisition unit 21 include the region of the subject H and the direct radiation region obtained by direct radiation exposure to the radiation detectors 5 and 6. The region of the subject H includes soft tissue and bone. Soft tissue components of the human body include muscle, fat, blood, and water. In this embodiment, non-fat tissue, which also contains blood and water, is treated as muscle.

[0138] The soft tissue regions of the first and second radiographic images G1 and G2 include only the soft tissue components of the subject H. The bone regions of the first and second radiographic images G1 and G2 are actually regions in which the bone components and the soft tissue components are mixed.

[0139] The following describes the processing performed by the region identification unit 31 , the characteristic derivation unit 32 , and the image derivation unit 33 . Figure 23 Schematically shows the process of bone image derivation according to the fourth embodiment. Figure 23 For convenience of explanation, it is assumed that the first radiographic image G1 and the second radiographic image G2 do not include a direct radiation region but include a rectangular bone region in a soft tissue region.

[0140] The region identification unit 31 identifies the bone region and the soft tissue region in the first radiographic image G1 or the second radiographic image G2. Therefore, the region identification unit 31 derives attenuation characteristics related to radiation attenuation in at least the region of the subject H in the first radiographic image G1 or the second radiographic image G2, and identifies the soft tissue region and the bone region based on the attenuation characteristics in the region of the subject H. The region identification unit 31 derives a first attenuation image CL and a second attenuation image CH representing the amount of radiation attenuation by the subject H from the first radiographic image G1 and the second radiographic image G2, respectively, and derives an attenuation ratio (i.e., the ratio between corresponding pixels in the first attenuation image CL and the second attenuation image CH) as the attenuation characteristic.

[0141] Here, the pixel values ​​of the first attenuation image CL represent the attenuation of low-energy radiation by the subject H, and the pixel values ​​of the second attenuation image CH represent the attenuation of high-energy radiation by the subject H. The first attenuation image CL and the second attenuation image CH are derived from the first radiographic image G1 and the second radiographic image G2 using the following equations (3) and (4). Gd1 in equation (3) represents the pixel value of the direct radiation region in the first radiographic image G1, and Gd2 in equation (4) represents the pixel value of the direct radiation region in the second radiographic image G2.

[0142] CL(x, y) = Gd1 - G1(x, y) (3)

[0143] CH(x, y) = Gd2 - G2(x, y) (4)

[0144] Next, the region identification unit 31 derives an attenuation ratio map representing the attenuation ratio of radiation between the first radiographic image G1 and the second radiographic image G2. Specifically, the attenuation ratio map M1 is derived by deriving the ratio between corresponding pixels in the first attenuation image CL and the second attenuation image CH using the following equation (5). The attenuation ratio is an example of a characteristic related to the attenuation of radiation.

[0145] M1(x,y)=CL(x,y) / CH(x,y) (5)

[0146] Here, in the first radiographic image G1 and the second radiographic image G2, the attenuation ratio of the region consisting solely of soft tissue components is lower than the attenuation ratio of the region consisting solely of bone components. Therefore, the region identification unit 31 compares the attenuation ratios of each pixel in the attenuation ratio map M1 and identifies the region consisting of pixels with attenuation ratios greater than a predetermined threshold as the bone region, and identifies the region excluding the bone region as the soft tissue region. Alternatively, the region identification unit 31 may compare the attenuation ratios of each pixel in the attenuation ratio map M1 with those of the surrounding pixels and identify pixels with a greater attenuation ratio than the surrounding pixels as pixels in the bone region.

[0147] The characteristics derivation unit 32 derives soft tissue characteristics related to radiation attenuation based on the first and second radiographic images G2 in a soft tissue region that includes only the first radiographic image G1 or the second radiographic image G2. Furthermore, the characteristics derivation unit 32 derives soft tissue characteristics in the skeletal region based on the soft tissue characteristics derived in the soft tissue region surrounding the skeletal region. In this embodiment, the characteristics derivation unit 32 derives the attenuation ratio between the first radiographic image G1 and the second radiographic image G2 as the soft tissue characteristic.

[0148] In the fourth embodiment, the region identification unit 31 derives the attenuation ratio, a characteristic related to radiation attenuation, in the region of the subject H in the first radiographic image G1 or the second radiographic image G2. Therefore, the characteristic derivation unit 32 uses the attenuation ratio of the soft tissue region in the attenuation ratio map M1 derived by the region identification unit 31 as the characteristic of the soft tissue in the soft tissue region. In other words, the region identification unit 31 also functions as the characteristic derivation unit 32.

[0149] On the other hand, regarding the skeletal region, the attenuation ratio, i.e., the soft tissue characteristic for the skeletal region, is derived by interpolating the attenuation ratios of the soft tissue regions surrounding the skeletal region. Alternatively, instead of interpolation, the median, mean, or a value (a predetermined ratio from the side with the smaller attenuation ratio) of the attenuation ratios of the soft tissue regions in the attenuation ratio map M1 can be used as the attenuation ratio for the skeletal region.

[0150] Thus, the characteristic derivation unit 32 deduces the characteristics of the soft tissue from the region of the subject H in the first radiographic image G1 or the second radiographic image G2. The soft tissue characteristic is the attenuation ratio of the soft tissue region. In the fourth embodiment, when the image derivation unit 33, described later, deduces a skeletal image, the deduced soft tissue characteristics are used as a soft tissue elimination coefficient K1 for eliminating the soft tissue.

[0151] The image derivation unit 33 derivates a skeleton image Gb in which the skeleton components are emphasized based on the characteristics of the soft tissue in the region of the subject H in the first radiographic image G1 or the second radiographic image G2 .

[0152] In the fourth embodiment, the image derivation unit 33 first derives an initial bone attenuation image in which bones are emphasized based on the first attenuation image CL, the second attenuation image CH, and the soft tissue elimination coefficient K1, which is a characteristic of soft tissue. Specifically, the image derivation unit 33 derives the initial bone attenuation image Cb0 using the following equation (6).

[0153] Cb0(x,y)=CL(x,y)-CH(x,y)×K1(x,y) (6)

[0154] The pixel values ​​in the bone region of the initial bone attenuation image Cb0 derived as described above represent the difference between: the value obtained by replacing the bone attenuation with the attenuation of the soft tissue, assuming the presence of soft tissue corresponding to the thickness of the bone; and the actual bone attenuation. Therefore, the image has a lower contrast than the bone attenuation image originally intended to be derived. If this low-contrast bone attenuation image is used, as described later, when the soft tissue attenuation image is derived by subtracting the bone attenuation image from the first attenuation image CL or the second attenuation image CH, the bone component cannot be effectively removed.

[0155] Therefore, the image derivation unit 33 derives the bone attenuation image Cb1 by aligning the contrast of the initial bone attenuation image Cb0 with the contrast of the first attenuation image CL or the second attenuation image CH. In the fourth embodiment, the contrast of the initial bone attenuation image Cb0 is aligned with the contrast of the first attenuation image CL. Therefore, the image derivation unit 33 converts the contrast of the initial bone attenuation image Cb0 by multiplying the initial bone attenuation image Cb0 by a contrast conversion coefficient. The image derivation unit 33 then derives the correlation between the difference image ΔCL derived by subtracting the contrast-converted initial bone attenuation image Cb0 from the first attenuation image CL and the initial bone attenuation image Cb0. The contrast conversion coefficient is then determined to minimize the correlation, and the determined contrast conversion coefficient is multiplied by the initial bone attenuation image Cb0 to derive the bone attenuation image Cb1.

[0156] Alternatively, a table may be pre-created that indicates contrast conversion coefficients that correlate the contrast of the initial bone attenuation image Cb0 with body thickness. In this case, the body thickness of the subject H may be derived through measurement or other means. Based on the contrast of the initial bone attenuation image Cb0 and the body thickness, contrast conversion coefficients may be derived from the table. The derived contrast conversion coefficients are then used to convert the initial bone attenuation image Cb0, thereby deriving the bone attenuation image Cb1.

[0157] Then, the image derivation unit 33 derivates the soft tissue attenuation image Cs1 by subtracting the bone attenuation image Cb1 from the first attenuation image CL using the following equation (7).

[0158] Cs1(x,y)=CL(x,y)-Cb1(x,y) (7)

[0159] Furthermore, the image derivation unit 33 derivates the skeleton image Gb using the following equation (8). Furthermore, when derivation of the soft tissue image Gs, the image derivation unit 33 derivates the soft tissue image Gs using the following equation (9).

[0160] Gb(x,y)=Gd1(x,y)-Cb1(x,y) (8)

[0161] Gs(x,y)=Gd2(x,y)-Cs1(x,y) (9)

[0162] Next, the skeleton image derivation process performed in the fourth embodiment will be described. Figure 24 This is a flowchart illustrating the skeletal image derivation process performed in the fourth embodiment. It is assumed that the acquisition of the first and second radiographic images G1 and G2 by the image acquisition unit 21 and the removal of scattered radiation components from the first and second radiographic images G1 and G2 by the scattered radiation removal unit 22 have completed. Therefore, in the following description, it is assumed that the scattered radiation components have been removed from the first and second radiographic images G1 and G2.

[0163] First, the region identification unit 31 derives a first attenuation image CL and a second attenuation image CH from the first radiographic image G1 and the second radiographic image G2 (deriving attenuation images: step ST11). The unit then identifies a soft tissue region (including only soft tissue components) and a bone region (including bone) in the first attenuation image CL or the second attenuation image CH (step ST12). Next, the characteristic derivation unit 32 derives the characteristics (attenuation ratio) of the soft tissue components related to the attenuation of the radiographic images in the soft tissue regions (step ST13). Next, the characteristic derivation unit 32 derives the characteristics of the soft tissue components in the bone regions (step ST14).

[0164] Next, the image derivation unit 33 derives the initial bone attenuation image Cb0 (step ST15), converts the contrast of the initial bone attenuation image Cb0, and thereby derives the bone attenuation image Cb1 (step ST16). In addition, the image derivation unit 33 derives the soft tissue attenuation image Cs1 by subtracting the bone attenuation image Cb1 from the first attenuation image CL (step ST17). Next, the image derivation unit 33 derives the bone image Gb using the above formula (8) (step ST18). After deriving the bone image Gb, the above-mentioned step Figure 10 The process of step ST4 of the flowchart shown in FIG.

[0165] In addition, in each of the above embodiments, the femur is used as the target bone, but the present invention is not limited thereto and the vertebrae may also be used as the target bone.

[0166] In particular, if osteoporosis causes a decrease in bone mineral content in the vertebrae, worsening the condition, the vertebrae are compressed vertically and deformed, potentially leading to compression fractures. Therefore, even with the vertebrae as the target bone, the present embodiment can still derive the probability of osteoporosis, P0.

[0167] Furthermore, in this embodiment, in addition to the femur and vertebrae, any bones such as the femur and tibia around the knee joint can be used as target bones.

[0168] Furthermore, in the above-described embodiments, scattered radiation components are removed from the first and second radiation images G1 and G2, but the present invention is not limited thereto. The bone image Gb may be derived without removing scattered radiation components from the first and second radiation images G1 and G2.

[0169] Furthermore, in the above-described embodiments, the first and second radiographic images G1 and G2 are acquired using a single exposure method during energy subtraction processing, but this is not limiting. Alternatively, the first and second radiographic images G1 and G2 may be acquired using a so-called double exposure method, in which two images are captured using only one radiation detector. With the double exposure method, the position of the subject H in the first and second radiographic images G1 and G2 may shift due to movement of the subject H. Therefore, the processing of this embodiment is preferably performed after the subject's position is aligned in the first and second radiographic images G1 and G2.

[0170] Furthermore, in each of the above-described embodiments, image processing is performed using radiographic images (acquired in a system that uses the first radiation detector 5 and the second radiation detector 6 to capture the first radiographic image G1 and the second radiographic image G2 of the subject H). However, the technology of the present invention can also be applied to situations where the first radiographic image G1 and the second radiographic image G2 are acquired using a storage fluorescent sheet in place of a radiation detector. In this case, the first radiographic image G1 and the second radiographic image G2 can be acquired by overlapping two storage fluorescent sheets, irradiating the subject H with radiation that has passed through the sheet, thereby storing and recording radiographic image information of the subject H on each storage fluorescent sheet, and photoelectrically reading the radiographic image information from each storage fluorescent sheet. Furthermore, when acquiring the first radiographic image G1 and the second radiographic image G2 using the storage fluorescent sheet, a double exposure method can also be employed.

[0171] Furthermore, the radiation in the above-mentioned embodiment is not particularly limited, and in addition to X-rays, α-rays, γ-rays, and the like can also be used.

[0172] Furthermore, in the above-described embodiment, various processors described below can be used as the hardware configuration of the processing units that perform various processes (e.g., the image acquisition unit 21, the scattered ray removal unit 22, the bone image derivation unit 23, the segmentation unit 24, the likelihood derivation unit 25, the display control unit 26, the region identification unit 31, the characteristic derivation unit 32, and the image derivation unit 33). As described above, these various processors include general-purpose processors (CPUs) that execute software (programs) and function as various processing units. These processors also include processors whose circuit configurations can be modified after manufacturing, such as programmable logic devices (PLDs) such as FPGAs (Field Programmable Gate Arrays), and processors with circuit configurations specifically designed to perform specific processes, such as dedicated electrical circuits such as ASICs (Application Specific Integrated Circuits).

[0173] A single processing unit may be composed of one of these various processors, or a combination of two or more processors of the same or different types (e.g., a combination of multiple FPGAs or a combination of a CPU and an FPGA). Furthermore, a single processor may constitute multiple processing units.

[0174] As examples of multiple processing units composed of a single processor, there are the following methods: For example, as exemplified by computers such as clients and servers, a single processor is composed of a combination of one or more CPUs and software, and this processor functions as multiple processing units. Another method is exemplified by a system-on-chip (SoC) system, which uses a processor that implements the functions of an entire system including multiple processing units on a single integrated circuit (IC). In this way, various processing units are constructed as hardware using one or more of the various processors described above.

[0175] Furthermore, as the hardware configuration of these various processors, more specifically, a circuit (Circuitry) formed by combining circuit elements such as semiconductor elements can be used.

[0176] The following describes additional notes of the present invention.

[0177] (Supplementary Note 1)

[0178] An image processing device comprising at least one processor,

[0179] The processor performs the following processing:

[0180] acquiring two radiographic images obtained by imaging a subject with radiation, wherein the radiation has different energy distributions and the subject includes bones and soft tissue;

[0181] deriving a bone image including the fine structure inside the bone based on the two radiographic images; and

[0182] The possibility of osteoporosis is derived based on the bone image or a partial image of the bone image including the target bone.

[0183] (Supplementary Note 2)

[0184] The image processing device according to Supplementary note 1, wherein

[0185] The bone image includes the trabecular structure of cancellous bone included in the bone.

[0186] (Supplementary Note 3)

[0187] The image processing device according to Supplementary note 2, wherein

[0188] The trabecular structure of the cancellous bone is depicted in the bone image.

[0189] (Supplementary Note 4)

[0190] The image processing device according to Supplementary note 3, wherein:

[0191] The bone image has a spatial resolution of less than 0.2 mm / pixel.

[0192] (Supplementary Note 5)

[0193] The image processing device according to any one of Supplementary Notes 1 to 4, wherein:

[0194] The processor extracts the partial image from at least one of the two radiographic images and the bone image.

[0195] (Supplementary Note 6)

[0196] The image processing device according to Supplementary note 5, wherein

[0197] The processor extracts at least one target bone from at least one of the two radiographic images and the bone image, and extracts the partial image based on the center of gravity of the target bone.

[0198] (Supplementary Note 7)

[0199] The image processing device according to any one of Supplementary Notes 1 to 6, wherein:

[0200] The processor functions as a trained model that has been subjected to machine learning so as to output the possibility of osteoporosis by inputting the bone image or the partial image.

[0201] (Supplementary Note 8)

[0202] The image processing device according to any one of Supplementary Notes 1 to 6, wherein:

[0203] The processor functions as a trained model that has been subjected to machine learning so as to output the possibility of osteoporosis by inputting imaging conditions of the two radiographic images in addition to the bone image or the partial image.

[0204] (Supplementary Note 9)

[0205] The image processing device according to any one of Supplementary Notes 1 to 6, wherein:

[0206] The processor derives the possibility of osteoporosis based on statistical values ​​of pixel values ​​of the target bone contained in the bone image or the partial image.

[0207] (Supplementary Note 10)

[0208] The image processing device according to Supplementary note 9, wherein

[0209] The statistical value is at least one of an average value, a standard deviation, a maximum value, and a minimum value of pixel values ​​in the target bone and each surrounding area of ​​the target bone.

[0210] (Supplementary Note 11)

[0211] The image processing apparatus according to Supplementary note 10, wherein:

[0212] The processor derives the possibility by performing weighted operations on a plurality of statistical values ​​including an average value, a standard deviation, a maximum value, and a minimum value of pixel values ​​in the target bone and each surrounding area of ​​the target bone.

[0213] (Supplementary Note 12)

[0214] The image processing device according to Supplementary note 11, wherein

[0215] The weights of the weighted operation are derived through machine learning.

[0216] (Supplementary Note 13)

[0217] The image processing device according to any one of Supplementary Notes 1 to 12, wherein:

[0218] The processor notifies the possibility.

[0219] (Supplementary Note 14)

[0220] The image processing device according to any one of Supplementary Notes 1 to 13, wherein:

[0221] When the probability is above a preset threshold, the processor issues a warning.

[0222] (Supplementary Note 15)

[0223] The image processing device according to any one of Supplementary Notes 1 to 14, wherein:

[0224] The processor determines a treatment priority for osteoporosis based on the likelihood and notifies the priority.

[0225] (Supplementary Note 16)

[0226] The image processing device according to any one of Supplementary Notes 1 to 14, wherein:

[0227] The processor removes scattered ray components included in the two radiographic images, thereby deriving the bone image.

[0228] (Supplementary Note 17)

[0229] The image processing device according to any one of Supplementary Notes 1 to 16, wherein:

[0230] The processor performs the following processing:

[0231] in a soft tissue region of the two radiographic images including only the soft tissue, deriving a characteristic of the soft tissue related to attenuation of the radiation based on the two radiographic images;

[0232] deriving the characteristics of the soft tissue in the bone region including the bone in the two radiographic images based on the characteristics of the soft tissue derived in the soft tissue region located around the bone region; and

[0233] The image in which the bones are emphasized is derived as the bone image based on the characteristics of the soft tissue in at least the region of the object in the two radiographic images.

[0234] (Supplementary Note 18)

[0235] The image processing device according to any one of Supplementary Notes 1 to 17, wherein:

[0236] The target bone is a femur or a vertebra.

[0237] (Supplementary Note 19)

[0238] An image processing method, wherein:

[0239] The computer performs the following processing:

[0240] acquiring two radiographic images obtained by imaging a subject with radiation, wherein the radiation has different energy distributions and the subject includes bones and soft tissue;

[0241] deriving a bone image including the fine structure inside the bone based on the two radiographic images; and

[0242] The possibility of osteoporosis is derived based on the bone image or a partial image of the bone image including the target bone.

[0243] (Supplementary Note 20)

[0244] An image processing program causes a computer to execute the following steps:

[0245] acquiring two radiographic images obtained by imaging a subject with radiation, wherein the radiation has different energy distributions and the subject includes bones and soft tissue;

[0246] deriving a bone image including the fine structure inside the bone based on the two radiographic images; and

[0247] The possibility of osteoporosis is derived based on the bone image or a partial image of the bone image including the target bone.

[0248] Explanation of symbols

[0249] 1- Imaging device, 3- Radiation source, 5, 6- Radiation detector, 7- Radiation energy conversion filter, 9- Image storage system, 10, 10A, 10B- Image processing device, 11- CPU, 12A- Image processing program, 13- Storage device, 14- Display, 15- Input device, 16- Memory, 17- Network I / F, 18- Bus, 21- Image acquisition unit, 22- Scattered ray removal unit, 23- Skeleton image derivation unit, 24- Segmentation unit, 25A, 25B- Trained model, 26- Display control unit, 31- Region determination unit, 32- Feature derivation unit, 33- Image derivation unit, 40, 40 A-training data, 41-learning data, 42-correct data, 43-pixel values ​​of the partial image including the target bone, 44-photographic conditions, 50, 50A, 50B-display screen, 51-possibility of osteoporosis, 52-warning, 53-treatment priority, 60-neural network, 61-input layer, 62-intermediate layer, 63-output layer, 65-convolutional layer, 66-pooling layer, 67-fully connected layer, A1-femur region, A2-pelvic region, A3-vertebral region, B(x, y)-bone mineral content of each pixel, G1-first radiographic image, G2-second radiographic image, Gb-bone image.

Claims

1. An image processing device comprising at least one processor, The processor performs the following processing: acquiring two radiographic images obtained by imaging a subject with radiation, wherein the radiation has different energy distributions and the subject includes bones and soft tissue; deriving a bone image including the fine structure inside the bone based on the two radiographic images; and The possibility of osteoporosis is derived based on the bone image or a partial image of the bone image including the target bone.

2. The image processing apparatus according to claim 1, wherein: The bone image includes the trabecular structure of cancellous bone included in the bone.

3. The image processing apparatus according to claim 2, wherein: The trabecular structure of the cancellous bone is depicted in the bone image.

4. The image processing apparatus according to claim 3, wherein: The bone image has a spatial resolution of less than 0.2 mm / pixel.

5. The image processing apparatus according to claim 1, wherein: The processor extracts the partial image from at least one of the two radiographic images and the bone image. The image processing apparatus according to claim 5 , wherein: The processor extracts at least one target bone from at least one of the two radiographic images and the bone image, and extracts the partial image based on the center of gravity of the target bone.

7. The image processing apparatus according to claim 1, wherein: The processor functions as a trained model that has been subjected to machine learning so as to output the possibility of osteoporosis by inputting the bone image or the partial image.

8. The image processing apparatus according to claim 1, wherein: The processor functions as a trained model that has been subjected to machine learning so as to output the possibility of osteoporosis by inputting imaging conditions of the two radiographic images in addition to the bone image or the partial image.

9. The image processing apparatus according to claim 1, wherein: The processor derives the possibility of osteoporosis based on statistical values ​​of pixel values ​​of the target bone contained in the bone image or the partial image.

10. The image processing apparatus according to claim 9, wherein: The statistical value is at least one of an average value, a standard deviation, a maximum value, and a minimum value of pixel values ​​in the target bone and each surrounding area of ​​the target bone. The image processing apparatus according to claim 10 , wherein: The processor derives the possibility by performing weighted operations on a plurality of statistical values ​​including an average value, a standard deviation, a maximum value, and a minimum value of pixel values ​​in the target bone and each surrounding area of ​​the target bone.

12. The image processing apparatus according to claim 11, wherein: The weights of the weighted operation are derived through machine learning.

13. The image processing apparatus according to claim 1, wherein: The processor notifies the possibility.

14. The image processing apparatus according to claim 1, wherein: When the probability is above a preset threshold, the processor issues a warning.

15. The image processing apparatus according to claim 1, wherein: The processor determines a treatment priority for osteoporosis based on the likelihood and notifies the priority.

16. The image processing apparatus according to claim 1, wherein: The processor removes scattered ray components included in the two radiographic images, thereby deriving the bone image.

17. The image processing apparatus according to claim 1, wherein: The processor performs the following processing: in a soft tissue region of the two radiographic images including only the soft tissue, deriving a characteristic of the soft tissue related to attenuation of the radiation based on the two radiographic images; deriving the characteristics of the soft tissue in the bone region including the bone in the two radiographic images based on the characteristics of the soft tissue derived in the soft tissue region located around the bone region; and The image in which the bones are emphasized is derived as the bone image based on the characteristics of the soft tissue in at least the region of the object in the two radiographic images.

18. The image processing apparatus according to claim 1, wherein: The target bone is a femur or a vertebra.

19. An image processing method, wherein: The computer performs the following processing: acquiring two radiographic images obtained by imaging a subject with radiation, wherein the radiation has different energy distributions and the subject includes bones and soft tissue; deriving a bone image including a fine structure inside the bone based on the two radiographic images; and The possibility of osteoporosis is derived based on the bone image or a partial image of the bone image including the target bone.

20. An image processing program causing a computer to execute the following steps: acquiring two radiographic images obtained by imaging a subject with radiation, wherein the radiation has different energy distributions and the subject includes bones and soft tissue; deriving a bone image including the fine structure inside the bone based on the two radiographic images; and The possibility of osteoporosis is derived based on the bone image or a partial image of the bone image including the target bone.

Images