Image processing method, image processing system, image processing program, and recording medium

Abstract

To more accurately correct blur in a radiation image without measuring the blur for each imaging system. [Solution] The image processing method includes an acquisition step (S01) of acquiring a radiation image obtained by imaging with an imaging system that uses radiation to capture images, as well as inference information, which is at least one of information related to the position of the object of the imaging in the imaging direction, information related to a radiation source included in the imaging system, and information related to a radiation detector included in the imaging system, and a correction step (S02) of inputting information based on the acquired radiation image and the inference information into an inference model generated by machine learning training, and correcting blurring of the radiation image.

Description

[Technical Field]

[0001] The present invention relates to an image processing method, an image processing system, an image processing program, and a recording medium for processing a radiation image. [Background technology]

[0002] Patent Document 1 discloses that the resolution of an X-ray image obtained by X-ray imaging is increased by using a blur function that indicates the blur of the focus in an X-ray imaging system. In Patent Document 1, the blur function that indicates the blur of the focus is an impulse response. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2016-224837 Summary of the Invention [Problem to be solved by the invention]

[0004] In Patent Document 1, the blur function is set by measuring the blur of the focal point of an X-ray imaging system. Usually, the focal blur information indicating the response to X-rays corresponds to the imaging system (X-ray imaging system) that performs imaging using X-rays. Therefore, in order to correct the blur of an X-ray image using the method disclosed in Patent Document 1, it is necessary to actually measure the blur for each imaging system that performed the imaging, and it is difficult to accurately correct the blur for an X-ray image obtained by imaging with an imaging system whose blur has not been actually measured. Furthermore, the same problem can occur with radiation images other than X-ray images.

[0005] The present invention has been made in view of the above, and aims to provide an image processing method, an image processing system, an image processing program, and a recording medium that can more accurately correct blur in radiographic images without having to measure the blur for each imaging system. [Means for solving the problem]

[0006] In order to achieve the above-mentioned object, the image processing method of the present invention includes an acquisition step of acquiring a radiological image obtained by imaging with an imaging system that performs imaging using radiation, as well as inference information that is at least any of information related to the position of the object of the imaging in the imaging direction, information related to a radiation source included in the imaging system, and information related to a radiation detector included in the imaging system; and a correction step of inputting information based on the radiological image and the inference information acquired in the acquisition step into an inference model generated by machine learning training, and correcting blurring of the radiological image.

[0007] In the image processing method according to the present invention, information based on the radiographic image and the inference information is input to an inference model generated by machine learning training, and blur in the radiographic image is corrected. Therefore, the image processing method according to the present invention can more accurately correct blur in the radiographic image without actually measuring blur for each imaging system.

[0008] Furthermore, in order to achieve the above-mentioned object, the image processing method according to the present invention includes an acquisition step of acquiring inference information, which is at least one of a radiation image obtained by imaging with an imaging system that performs imaging using radiation, and information related to the position of an object to be imaged in the imaging direction, information related to a radiation source included in the imaging system, and information related to a radiation detector included in the imaging system; and a blur information generation step of inputting information based on the radiation image and the inference information obtained in the acquisition step into an inference model generated by machine learning training, to generate blur information indicating a response to radiation in the radiation image obtained by imaging with the imaging system.

[0009] In the image processing method according to the present invention, information based on a radiographic image and inference information is input to an inference model generated by machine learning training, and blur information for a radiographic image obtained by imaging with an imaging system is generated. The generated blur information can be used to correct blur in the radiographic image obtained by imaging with the imaging system. Therefore, the image processing method according to the present invention makes it possible to correct blur in a radiographic image without actually measuring blur for each imaging system.

[0010] The information relating to the position of the object in the imaging direction may be a magnification rate according to the position. With this configuration, blurring of the radiographic image can be corrected more appropriately.

[0011] The information relating to the radiation source included in the imaging system may be information indicating the size of the focal point of the radiation source. With this configuration, blurring of the radiographic image can be corrected more appropriately.

[0012] The information related to the radiation detector included in the imaging system may be at least one of information related to the scintillator included in the radiation detector and information indicating the size of the pixels. With this configuration, blurring of the radiographic image can be corrected more appropriately.

[0013] In addition to being described as an invention of an image processing method as described above, the present invention can also be described as an invention of an image processing system, an image processing program, and a recording medium as described below. These are essentially the same invention, just in different categories, and have similar functions and effects.

[0014] That is, the image processing system according to the present invention comprises an acquisition means for acquiring inference information, which is at least one of a radiological image obtained by imaging with an imaging system that uses radiation and the position of the object being imaged in the imaging direction, information relating to a radiation source included in the imaging system, and information relating to a radiation detector included in the imaging system, and a correction means for inputting information based on the radiological image and the inference information acquired by the acquisition means into an inference model generated by machine learning training, and correcting blurring of the radiological image.

[0015] The image processing system according to the present invention also includes an acquisition means for acquiring inference information, which is at least one of a radiation image obtained by imaging with an imaging system that uses radiation and information related to the position of an object to be imaged in the imaging direction, information related to a radiation source included in the imaging system, and information related to a radiation detector included in the imaging system; and a blur information generation means for inputting information based on the radiation image and the inference information obtained by the acquisition means into an inference model generated by machine learning training, and generating blur information indicating a response to radiation in the radiation image obtained by imaging with the imaging system.

[0016] Furthermore, the image processing program of the present invention causes a computer to function as: an acquisition means for acquiring inference information, which is at least one of a radiological image obtained by imaging with an imaging system that uses radiation and the position of the object to be imaged in the imaging direction, information related to a radiation source included in the imaging system, and information related to a radiation detector included in the imaging system; and a correction means for inputting information based on the radiological image and the inference information acquired by the acquisition means into an inference model generated by machine learning training, and correcting blurring of the radiological image.

[0017] Furthermore, the image processing program of the present invention causes a computer to function as: an acquisition means for acquiring inference information, which is at least one of a radiological image obtained by imaging with an imaging system that performs imaging using radiation, and information related to the position of the object of the imaging in the imaging direction, information related to the radiation source included in the imaging system, and information related to the radiation detector included in the imaging system; and a blur information generation means for inputting information based on the radiological image and the inference information obtained by the acquisition means into an inference model generated by machine learning training, and generating blur information indicating the response to radiation in the radiological image obtained by imaging with the imaging system.

[0018] A recording medium according to the present invention is a computer-readable recording medium on which the image processing program described above is recorded. [Effects of the Invention]

[0019] According to the present invention, blur in a radiographic image can be corrected more accurately without measuring the blur for each imaging system. [Brief explanation of the drawings]

[0020] [Figure 1] 1 is a diagram showing the configuration of an image processing system according to a first embodiment of the present invention. [Figure 2] FIG. 2 is a diagram schematically illustrating a configuration of a part of a detector included in the imaging system. [Figure 3] FIG. 1 is a diagram illustrating an inference model used in an image processing system. [Figure 4] A figure showing an example of an image used to generate an inference model. [Figure 5] FIG. 10 is a diagram showing an example of a PSF that is out-of-focus information. [Figure 6] A figure showing an example of generating images used to generate an inference model. [Figure 7] 3 is a flowchart showing an image processing method which is processing executed in the image processing system according to the first embodiment of the present invention. [Figure 8] 5A to 5C are diagrams illustrating an example of blur correction performed in the first embodiment. [Figure 9] 10A to 10C are diagrams illustrating an example of blur correction performed by changing the inference information in the first embodiment. [Figure 10] 10A to 10C are diagrams illustrating an example of blur correction performed by changing the inference information in the first embodiment. [Figure 11] 10A to 10C are diagrams illustrating an example of blur correction performed by changing the inference information in the first embodiment. [Figure 12] 10A to 10C are diagrams illustrating an example of blur correction performed by changing the inference information in the first embodiment. [Figure 13] 10A to 10C are diagrams illustrating an example of blur correction performed by changing the inference information in the first embodiment. [Figure 14] 10A to 10C are diagrams illustrating an example of blur correction performed by changing the inference information in the first embodiment. [Figure 15] 10A to 10C are diagrams illustrating an example of blur correction performed by changing the inference information in the first embodiment. [Figure 16] 10A to 10C are diagrams illustrating an example of blur correction performed by changing the inference information in the first embodiment. [Figure 17] 1 is a diagram showing the configuration of an image processing program according to a first embodiment of the present invention, together with a recording medium. [Figure 18] FIG. 10 is a diagram showing the configuration of an image processing system according to a second embodiment of the present invention. [Figure 19] FIG. 1 is a diagram illustrating an inference model used in an image processing system. [Figure 20] 10 is a flowchart showing an image processing method which is processing executed in an image processing system according to a second embodiment of the present invention. [Figure 21] FIG. 10 is a diagram showing the configuration of an image processing program according to a second embodiment of the present invention, together with a recording medium. DETAILED DESCRIPTION OF THE INVENTION

[0021] Hereinafter, an embodiment of an image processing method, an image processing system, an image processing program, and a recording medium according to the present invention will be described in detail with reference to the drawings. In the description of the drawings, the same elements are designated by the same reference numerals, and duplicated explanations will be omitted.

[0022] First Embodiment FIG. 1 shows an image processing system 10a according to this embodiment. The image processing system 10a is a system (device) that performs information processing for processing an image obtained by imaging using an imaging system 20. The imaging system 20 is a system (device) that captures images using radiation. In this embodiment, the imaging system 20 is an X-ray imaging system (X-ray camera) that captures images using X-rays. The radiation used for imaging by the imaging system 20 may be radiation other than X-rays. When radiation other than X-rays is used, X-rays can be read as radiation other than X-rays in the following description. The imaging system 20 may be the same as a conventional system.

[0023] The imaging system 20 obtains an X-ray image (radiation image) by irradiating an object to be imaged with X-rays and detecting the X-rays that have passed through the object. As shown in FIG. 1, the imaging system 20 includes an X-ray source 21 and a detector (radiation detector) 22. An object 30 to be imaged is positioned and placed between the X-ray source 21 and the detector 22 before imaging. The X-ray source 21 generates X-rays to be irradiated onto the object 30. The X-rays are generated from an X-ray focus of the X-ray source 21. The detector 22 is a device that detects the X-rays that have been irradiated from the X-ray source 21 toward the object 30 and passed through the object 30 to generate an image. The detector 22 includes a scintillator and a photodetector, which detect the X-rays and generate an image.

[0024] An X-ray image (radiographic image) obtained by imaging with the imaging system 20 will have blurring due to the X-ray focus of the X-ray source 21. Hereinafter, blurring due to the X-ray focus will be referred to as focal blurring. In addition, X-ray images may also have blurring due to factors other than the X-ray focus. For example, X-ray images may have blurring due to the scintillator included in the detector 22 of the imaging system 20. Hereinafter, blurring due to the scintillator will be referred to as scintillator blurring.

[0025] The image processing system 10a corrects blurring that occurs in an X-ray image obtained by imaging by the imaging system 20, i.e., performs processing to reduce or remove blurring in the X-ray image. Specifically, the image processing system 10a corrects blurring that occurs in an X-ray image and generates a corrected X-ray image. That is, the image processing system 10a performs automatic restoration correction of the X-ray image. For example, the image processing system 10a corrects out-of-focus blurring that occurs in an X-ray image. The image processing system 10a may also correct blurring caused by factors other than out-of-focus blurring. The image processing system 10a may also correct blurring that includes out-of-focus blurring and blurring caused by other factors.

[0026] The image processing system 10a is configured to include a conventional computer including a processor such as a CPU (Central Processing Unit), a memory, a communication module, and other hardware. The functions of the image processing system 10a, which will be described later, are realized by these components operating according to programs, etc. The computer that constitutes the image processing system 10a may be a computer system including multiple computers. The computer may also be configured using cloud computing or edge computing.

[0027] Next, the functions of the image processing system 10a according to this embodiment will be described. As shown in Fig. 1, the image processing system 10a includes an acquisition unit 11a and a correction unit 12a.

[0028] The acquisition unit 11a is an acquisition means for acquiring information for inference, which is at least one of an X-ray image obtained by imaging with an imaging system 20 that performs imaging using X-rays, a position of an object of the imaging in the imaging direction, information related to an X-ray source 21 included in the imaging system 20, and information related to a detector 22 included in the imaging system 20. The information related to the position of an object of the imaging in the imaging direction may be a magnification ratio according to the position. The information related to the X-ray source 21 included in the imaging system 20 may be information indicating the size of the focal point of the X-ray source 21. The information related to the detector 22 included in the imaging system 20 may be at least one of information related to a scintillator included in a radiation detector and information indicating the size of a pixel.

[0029] The acquisition unit 11a acquires an X-ray image and information, for example, as follows. The acquisition unit 11a acquires an X-ray image, which is a radiographic image to be subjected to blur correction. The X-ray image is obtained by imaging using the imaging system 20, and is an image of a preset size. The acquisition of the X-ray image is performed, for example, by receiving an X-ray image transmitted from the imaging system 20 or by accepting an input operation from a user. The acquisition of the X-ray image may also be performed by other methods.

[0030] The acquiring unit 11a also acquires inference information related to the acquired X-ray image. The inference information is information related to blurring of the X-ray image and is information used to correct the X-ray image. For example, the inference information is information related to the imaging system 20 and imaging conditions. The acquiring unit 11a acquires, as the inference information, information related to the position of the object 30 in the imaging direction when capturing the X-ray image, information related to the X-ray source 21 included in the imaging system 20, and information related to the detector 22 included in the imaging system 20. The inference information is acquired, for example, by receiving an input operation from the user or by receiving information transmitted from the imaging system 20. The inference information may also be acquired by other methods.

[0031] The information relating to the position of the object 30 in the imaging direction when capturing an X-ray image is, for example, a magnification factor corresponding to the position. The magnification factor of the X-ray image corresponds to the position of the object 30 in the imaging direction. The defocus that occurs in the X-ray image corresponds to the magnification factor. For example, the acquisition unit 11a acquires the value of the magnification factor of the X-ray image as the information.

[0032] The magnification ratio M can be calculated, for example, by the following formula: Note that the object 30 to be imaged is assumed to be located on a straight line connecting the X-ray focal point of the X-ray source 21 and the position at which the detector 22 detects the X-rays. M=1+b / a In the above formula, a is the distance between the X-ray focal point of the X-ray source 21 and the object 30 to be imaged (FOD: Focus Object Distance). b is the distance between the object 30 to be imaged and the position where the X-rays are detected on the detector 22. Note that a+b is the distance between the X-ray focal point of the X-ray source 21 and the position where the X-rays are detected on the detector 22 (FDD: Focus Detector Distance).

[0033] The acquisition unit 11a may store a formula for calculating the magnification ratio in advance, acquire information for calculating the magnification ratio (the values ​​of a and b in the above example), and calculate and acquire the magnification ratio from the information. The magnification ratio M may also be a value corresponding to the position in the X-ray image. The acquisition unit 11a may acquire information other than the magnification ratio as information related to the position of the object 30 in the imaging direction in capturing the X-ray image.

[0034] The information related to the X-ray source 21 included in the imaging system 20 is, for example, information indicating the size of the focal spot of the X-ray source 21. The defocus occurring in the X-ray image depends on the size of the focal spot of the X-ray source 21.

[0035] The size of the focal spot is, for example, the size (for example, a value in μm units) of the focal spot of the X-ray source 21 when viewed from the imaging direction (the direction in which X-rays are generated). The size of the focal spot may be a value based on a criterion set in advance depending on the type of shape of the focal spot. Types of focal spot shapes include, for example, a Gaussian shape and a rectangular shape. A Gaussian-shaped focus is a focus in which the intensity distribution of the generated X-rays for each position is a Gaussian distribution. A rectangular-shaped focus is a focus in which the intensity of the generated X-rays for each position is uniform and the focus is rectangular.

[0036] In the case of a Gaussian-shaped focus, the size of the focus is the full width at half maximum of the Gaussian distribution. In the case of a rectangular focus, the size of the focus is the width of the rectangle (the length of one side of the rectangle). The size of the focus may be determined from the focus size mode (e.g., small focus mode, medium focus mode, or large focus mode) set when the X-ray source 21 is used. The acquisition unit 11a may acquire information other than information indicating the size of the focus of the X-ray source 21 as information related to the X-ray source 21 included in the imaging system 20.

[0037] The information related to the detector 22 included in the imaging system 20 is, for example, information related to the scintillator included in the detector 22 and information indicating the size of the pixels. The blur that occurs in the X-ray image depends on the size of the scintillator and the pixel. For example, the acquisition unit 11a acquires the thickness value of the scintillator (for example, a value in μm units) as the information related to the scintillator. Furthermore, the information related to the scintillator may be something other than the thickness value (for example, information indicating the material of the scintillator).

[0038] Furthermore, the acquiring unit 11a acquires a pixel size value (for example, a value in μm units) as information indicating the pixel size. As schematically shown in FIG. 2, the detector 22 includes a plurality of photodetectors (photodetecting elements) 22a corresponding to the respective pixels of the image generated by the imaging system 20. The photodetectors (photodetecting elements) 22a are arranged side by side on a plane perpendicular to the imaging direction (the direction in which X-rays are generated). As shown in FIG. 2, the pixel size is the size of the light-detecting portion of the photodetector 22a. The acquiring unit 11a may acquire information other than information related to the scintillator included in the detector 22 and information indicating the pixel size as information related to the detector 22 included in the imaging system 20.

[0039] The acquiring unit 11a does not need to acquire all of the above information as the inference information, but may acquire only a part of the above information. Also, the acquiring unit 11a may acquire information other than the above as the inference information.

[0040] For example, information indicating the shape of the focal point of the X-ray source 21 may be used as the information for inference. The out-of-focus blur that occurs in the X-ray image corresponds to the shape of the focal point of the X-ray source 21. The shape of the focal point is, for example, the two-dimensional shape of the position where the X-rays are generated on a plane perpendicular to the X-rays generated from the X-ray source 21 and the intensity of the X-rays for each position. The information indicating the shape of the focal point is information indicating the type of the shape of the focal point.

[0041] Furthermore, the information indicating the shape of the focal spot may include information indicating the intensity of the generated X-rays at each position. That is, the information indicating the shape of the focal spot may be a two-dimensional spatial intensity distribution of the X-ray source 21. The information indicating the shape of the focal spot of the X-ray source 21 can be obtained by a conventional measurement method (e.g., a pinhole camera method, a parallel pattern camera method, or a resolution method). Alternatively, the information indicating the shape of the focal spot of the X-ray source 21 can also be obtained by a conventional simulation method.

[0042] Furthermore, the size of the penumbra in the image captured by the imaging system 20 may be used as information for inference. The value of the size H of the penumbra can be calculated from the value of the magnification ratio M and the value of the size (magnification) f of the focal point of the X-ray source 21 using the following formula: H=(M-1)×f

[0043] The information related to the X-ray source 21 may be acquired based on information identifying the X-ray source 21. In this case, the acquisition unit 11 acquires the information identifying the X-ray source 21. The information identifying the X-ray source 21 is, for example, a model name of the X-ray source 21 that is set in advance for each type of X-ray source 21. The information identifying the X-ray source 21 is acquired by, for example, accepting an input operation from the user or receiving information transmitted from the imaging system 20. Alternatively, the information identifying the X-ray source 21 may be acquired by other methods.

[0044] The acquiring unit 11a stores in advance information indicating a correspondence relationship between information specifying the X-ray source 21 and the above-mentioned information related to the X-ray source 21. The acquiring unit 11a acquires information indicating the shape of the focal point of the X-ray source 21 that is associated with the acquired information specifying the X-ray source 21 in the correspondence relationship. The acquiring unit 11a may also acquire the above information to be used in subsequent processing based on information at the time of imaging (for example, the above-mentioned focal spot size mode, or tube voltage or tube current related to the X-ray source 21) as information related to the X-ray source 21. In addition to the above information related to the X-ray source 21, the acquiring unit 11a may also acquire the above information to be used in subsequent processing based on information that identifies an object related to the information (for example, the detector 22).

[0045] The acquisition unit 11a outputs the acquired X-ray image and inference information to the correction unit 12a.

[0046] The correction unit 12a is a correction means that inputs information based on the X-ray image and inference information acquired by the acquisition unit 11a into an inference model generated by machine learning training, and corrects blurring of the X-ray image.

[0047] The correction unit 12a corrects blur in an X-ray image using an inference model generated by machine learning training. The inference model includes, for example, a neural network. The neural network may be multi-layered. That is, the inference model may be generated by deep learning. The neural network may be a convolutional neural network (CNN). The neural network may be multi-modal. For example, the neural network may input both an image (image information) and a parameter or condition (text information) as shown below. The format of the inference model may be similar to that of an inference model generated by conventional machine learning training.

[0048] For example, as shown in FIG. 3, the inference model 30a receives an X-ray image 31 and information based on the inference information, and outputs a blur-corrected X-ray image 32. In this case, the inference model 30a has an input layer provided with neurons for inputting the X-ray image 31 and information based on the inference information. For example, the information input to the inference model 30a is the pixel value (brightness value) of each pixel of the X-ray image 31 and each value of the inference information. In this case, the input layer has neurons equal in number to the sum of the number of pixels in the X-ray image 31 and the number of pieces of inference information, and a corresponding value is input to each neuron. For example, as shown in FIG. 3, the inference model 30a has input channels for each type of information. In the example shown in FIG. 3, channel 1 is the X-ray image 31, channel 2 is the focal spot size, channel 3 is the magnification, channel 4 is the pixel size, and channel 5 is the scintillator thickness.

[0049] Furthermore, the inference model 30a is provided with neurons in the output layer for outputting the blur-corrected X-ray image 32. For example, the information output from the inference model is the pixel value of each pixel in the blur-corrected X-ray image. In this case, the output layer is provided with neurons equal to the number of pixels in the X-ray image, and each neuron outputs the pixel value of the corresponding pixel. For example, as shown in FIG. 3, the inference model 30a is provided with a channel (1ch) for outputting the blur-corrected X-ray image 32.

[0050] The inference model may be anything other than one that outputs a blur-corrected X-ray image, as long as it is used to obtain a blur-corrected X-ray image from information based on the X-ray image and the inference information.

[0051] The inference model functions to input information based on X-ray images and inference information, perform calculations according to the input, and output information. The inference model is expected to be used as a program module that is part of artificial intelligence software. The inference model is used, for example, in a computer having a processor and memory, and the computer's processor operates according to instructions from the model stored in the memory. For example, the computer's processor operates according to the instructions to input information to the model, perform calculations according to the model, and output results from the model. Specifically, the computer's processor operates according to the instructions to input information to the input layer of a neural network, perform calculations based on parameters such as learning weighting coefficients in the neural network, and output results from the output layer of the neural network. Note that the inference model may be configured using something other than a neural network.

[0052] The inference model used by the correction unit 12a is generated in advance by machine learning training and stored in the correction unit 12a. Generation of the inference model (machine learning training) will be described. To generate the inference model, a combination of learning images and inference information (learning data, teacher data) is prepared. The learning images are images that include blur when captured by an imaging system and corresponding images that do not include blur (clear images). The imaging system related to the learning images may be the imaging system 20 that captures images to obtain X-ray images to be corrected for blur, or it may not be the imaging system 20. The learning images are images of the same size as the X-ray images to be corrected by the image processing system 10a.

[0053] As the image including blur and the image without blur, for example, an X-ray image as shown in Fig. 4(a) may be used. When an X-ray image is used, an image captured by an imaging system related to the blur included in the learning image may be used, or an image captured by another imaging system may be used.

[0054] Furthermore, images other than X-ray images may be used as the blurred and non-blurred images. For example, images generated by simulation such as those shown in Figures 4(b) and 4(c) may be used as the blurred and non-blurred images. Alternatively, images such as natural images captured by a camera using visible light, such as those shown in Figure 4(d), may be used as the blurred and non-blurred images.

[0055] The blurred image and the blur-free image are prepared, for example, as follows: First, an image without blur is prepared. Then, blur information indicating a response to X-rays that causes blur in an X-ray image obtained by imaging using an imaging system related to the blur contained in the learning image is prepared. The blur information corresponds to inference information for learning. Using the blur information, an image with blur is generated from the image without blur.

[0056] For example, the blur information is out-of-focus information that indicates the response to X-rays according to the focus of the X-rays of the X-ray source of the imaging system. The out-of-focus information is, for example, a PSF (Point Spread Function). The PSF indicates how much a signal at a certain point (pixel) on an image spreads. The PSF is, for example, a two-dimensional spatial intensity distribution.

[0057] Figure 5 shows an example of a PSF. Figures 5(a)(b), 5(c)(d), 5(e)(f), and 5(g)(h) each show one PSF. Figures 5(a), 5(c), 5(e), and 5(g) show two-dimensional intensity, with the vertical and horizontal axes corresponding to pixel positions in the image. Figures 5(b), 5(d), 5(f), and 5(h) are graphs in which the horizontal axis indicates the pixel position on a line passing through the center of the PSF, and the vertical axis indicates the intensity of the response at that pixel position. The center position of each image and graph in Figure 5 is the reference point for the signal spread (the "point (pixel) on the image" mentioned above).

[0058] Figures 5(a) and 5(b) show PSFs for a focal spot size of 100 μm, a pixel size of 100 μm, a magnification of 1.1 times, and a rectangular focal spot shape (a shape in which the spatial intensity distribution is a rectangular distribution). Figures 5(c) and 5(d) show PSFs for a focal spot size of 100 μm, a pixel size of 100 μm, a magnification of 1.1 times, and a Gaussian focal spot shape (a shape in which the spatial intensity distribution is a Gaussian distribution). Figures 5(e) and 5(f) show PSFs for a focal spot size of 100 μm, a pixel size of 100 μm, a magnification of 4 times, and a rectangular focal spot shape. Figures 5(g) and 5(h) show PSFs for a focal spot size of 100 μm, a pixel size of 100 μm, a magnification of 4 times, and a Gaussian focal spot shape.

[0059] The out-of-focus information may be estimated (specified) by, for example, performing imaging (actual measurement) for specifying the out-of-focus information using an imaging system related to the blur contained in the learning image, or by calculation based on parameters at the time of imaging. The out-of-focus information may be estimated (specified) by a conventional method.

[0060] The blur information is scintillator blur information that indicates a response to X-rays according to a scintillator included in the detector 22 of the imaging system 20. The response to X-rays according to the scintillator causes scintillator blur in the X-ray image. The scintillator blur information is, for example, a PSF.

[0061] The scintillator blur information may be estimated (specified) by performing imaging (actual measurement) for specifying the scintillator blur information, or by performing simulation calculations based on parameters related to the scintillator at the time of imaging. The scintillator blur information may be estimated (specified) by a conventional method.

[0062] For example, the actual measurement method is performed as follows. First, the MTF (Modulation Transfer Function) is measured using the edge method or the chart method. The edge method involves capturing an oblique image of a tungsten chart and calculating the MTF. The chart method involves capturing an image of a square wave chart and calculating the MTF. Next, the MTF is subjected to an inverse Fourier transform to calculate the PSF, which is scintillator blur information, as spatial information.

[0063] The simulation method is performed as follows. First, the tube voltage, filter conditions, and scintillator type and thickness during simulation imaging are set. A Monte Carlo simulation is performed under these conditions. For example, a scintillator model is created and a light scattering simulation is performed. The PSF is calculated from the simulation results.

[0064] The PSF, which is the blur information obtained as described above, is convoluted with an image that does not contain blur to generate a blurred image. The generation of a blurred image from the blur information and the image that does not contain blur can be performed using conventional technology. Figure 6 shows an example of generating a blurred image 41 from an image that does not contain blur 42 and a PSF 43, which is the blur information.

[0065] Note that the blurred and non-blurred images may be prepared by methods other than those described above. The number of combinations of training images and inference information to be prepared is a number sufficient to adequately train the inference model.

[0066] The inference model is trained as follows using the prepared training images and inference information. The inference model is trained for each combination of training images and inference information. For example, information based on the training images containing blur and the inference information is input to the inference model, and a calculation is performed according to the inference model to obtain a corrected image. The corrected image obtained is compared with the training images that do not contain blur, and the parameters of the inference model are updated by error backpropagation based on the loss resulting from the comparison. In this case, when the training images containing blur are input to the inference model, the inference model is trained so that the inference model outputs the training images that do not contain blur.

[0067] The inference model may be generated by a method other than the above. The inference model may be generated in the image processing system 10a or by a system other than the image processing system 10a.

[0068] The correction unit 12a corrects blur in the X-ray image, for example, as follows: The correction unit 12a inputs the X-ray image and inference information from the acquisition unit 11a. The correction unit 12a inputs information based on the input X-ray image and inference information into an inference model to generate an X-ray image (reconstructed image) after blur correction. For example, if the inference model outputs an X-ray image after blur correction, the correction unit 12a acquires the output from the inference model as the X-ray image after blur correction. Furthermore, the correction unit 12a may correct blur in the X-ray image by a method other than the above, as long as it uses an inference model.

[0069] The correction unit 12a outputs the generated blur-corrected X-ray image. The blur-corrected X-ray image may be output in the same manner as in the conventional method depending on the intended use of the X-ray image. The above is the function of the image processing system 10a according to this embodiment.

[0070] Next, an image processing method, which is a process executed by the image processing system 10a according to this embodiment (an operating method performed by the image processing system 10a), will be described using the flowchart of Fig. 7. In this process, the acquisition unit 11a acquires an X-ray image and inference information obtained by imaging with the imaging system 20 (S01, acquisition step). The inference information is at least one of the position of the object in the imaging direction, information related to the X-ray source 21 included in the imaging system 20, and information related to the radiation detector included in the imaging system 20.

[0071] Next, the correction unit 12a inputs information based on the X-ray image and the inference information into the inference model, and corrects the blur of the X-ray image (S02, correction step). The X-ray image after blur correction is output from the correction unit 12a to a predetermined output destination (S03). The above is the image processing method, which is processing executed by the image processing system 10a according to this embodiment.

[0072] In this embodiment, information based on the X-ray image and the inference information is input to an inference model generated by machine learning training, and the blur of the X-ray image is corrected. Therefore, according to this embodiment, the blur of the X-ray image can be corrected more accurately without actually measuring the blur for each imaging system 20.

[0073] As in the present embodiment, the information relating to the position of the object to be imaged in the imaging direction as the inference information may be a magnification factor corresponding to the position. With this configuration, it is possible to more appropriately correct the blur of the X-ray image. However, the information relating to the position of the object to be imaged in the imaging direction may be something other than a magnification factor corresponding to the position.

[0074] As in the present embodiment, the information related to the X-ray source 21 included in the imaging system 20 may be information indicating the size of the focal spot of the X-ray source 21. With this configuration, blurring of the X-ray image can be corrected more appropriately. However, the information related to the X-ray source 21 may be information other than information indicating the size of the focal spot of the X-ray source 21.

[0075] As in the present embodiment, the information related to the detector 22 included in the imaging system 20 may be at least one of information related to the scintillator included in the detector 22 and information indicating the pixel size. With this configuration, blurring of the X-ray image can be corrected more appropriately. However, the information related to the X-ray source 21 and the information related to the detector 22 may be other than at least one of information related to the scintillator included in the detector 22 and information indicating the pixel size.

[0076] Next, an example according to this embodiment will be shown. As shown in Fig. 8, in this example, an image (original image) that does not contain blur is prepared, and the blur that occurs when the image is captured by the imaging system 20 is superimposed on the original image to generate an image that contains blur (composite blur image). The composite blur image is used as an image to be corrected for blur by the image processing system 10a, and a blur-corrected image (image restoration) is generated by the image processing system 10a.

[0077] The lower part of Fig. 8 is a graph corresponding to the image in the upper part. In this graph, the horizontal axis indicates the pixel position on a specific (identical) straight line in the image, and the vertical axis indicates the pixel value (brightness value) at that pixel position. Furthermore, the descriptions 2ch to 5ch on the left side of Fig. 8 are inference information corresponding to the conditions used when generating a composite blurred image from the original image. The descriptions 2ch to 5ch on the right side of Fig. 8 are inference information used (i.e., input to the inference model) when the image processing system 10a generates a blur-corrected image (image restoration) from the composite blurred image. The images, graphs, and descriptions in Figs. 9 to 16 are the same as those described above.

[0078] In the example shown here, the closer the blur-corrected image (image restoration) is to the blur-free image (original image), the more accurately the blur can be corrected. As shown in Fig. 8, according to this embodiment, the image processing system 10a functions well and can correct blur more accurately.

[0079] In the example shown in Fig. 8, the inference information corresponding to the conditions used when generating a composite blurred image from an original image (description of 2ch to 5ch on the left side of Fig. 8) is the same as the inference information used to generate a blur-corrected image (image restoration) by the image processing system 10a (description of 2ch to 5ch on the right side of Fig. 8) (i.e., input to the inference model). Below, an example will be shown in which the inference information used to generate a blur-corrected image (image restoration) is different from the inference information corresponding to the conditions used when generating a composite blurred image from an original image.

[0080] FIG. 9 shows an example in which the focal spot size (2ch) is set to 100 μm, instead of the original 300 μm. In this case, the image after blur correction (image restoration) is not properly restored. FIG. 10 shows an example in which the focal spot size (2ch) is set to 500 μm, instead of the original 300 μm. In this case, the image after blur correction (image restoration) is excessively restored.

[0081] FIG. 11 shows an example where the pixel size (4ch) is originally 100 μm but is changed to 300 μm. In this case, the image after blur correction (image restoration) is not properly restored. FIG. 12 shows an example where the pixel size (4ch) is originally 100 μm but is changed to 50 μm. In this case, the image after blur correction (image restoration) is excessively restored.

[0082] FIG. 13 shows an example where the magnification ratio (3ch) is set to 1x instead of the original 2x. In this case, the image after blur correction (image restoration) is not properly restored. FIG. 14 shows an example where the magnification ratio (3ch) is set to 3x instead of the original 2x. In this case, the image after blur correction (image restoration) is excessively restored.

[0083] FIG. 15 shows an example in which the thickness of the scintillator (5ch) is changed from the original 300 μm to 100 μm. In this case, the image after blur correction (image restoration) shows that the blur corresponding to the scintillator is not properly restored. However, the blur corresponding to the X-ray source is restored to some extent. FIG. 16 shows an example in which the thickness of the scintillator (5ch) is changed from the original 300 μm to 500 μm. In this case, the image after blur correction (image restoration) shows that excessive restoration has been performed.

[0084] As shown in the examples of Figures 9 to 16, unless appropriate inference information is used for blur correction, accurate blur correction cannot be performed. By using appropriate inference information for blur correction, accurate blur correction can be performed. The above are examples according to this embodiment.

[0085] Next, an image processing program for executing a series of processes by the image processing system 10a described above will be described. As shown in Fig. 17, the image processing program 100a is stored in a program storage area 111a formed on a computer-readable recording medium 110a that is inserted into a computer and accessed, or that is provided in the computer. The recording medium 110a may be a non-transitory recording medium.

[0086] The image processing program 100a includes an acquisition module 101a and a correction module 102a. The functions realized by executing the acquisition module 101a and the correction module 102a are similar to the functions of the acquisition unit 11a and the correction unit 12a of the image processing system 10a described above, respectively.

[0087] The image processing program 100a may be configured so that a part or all of it is transmitted via a transmission medium such as a communication line, and is received and recorded (including installed) by another device. Furthermore, each module of the image processing program 100a may be installed not on one computer but on one of multiple computers. In this case, the above-described series of processes are performed by a computer system consisting of the multiple computers.

[0088] Second Embodiment 18 shows an image processing system 10b according to this embodiment. Note that the points of this embodiment that will not be described below may be the same as those of the first embodiment unless there is a contradiction. The image processing system 10b is a system (device) that performs information processing for processing images obtained by imaging using the imaging system 20.

[0089] The image processing system 10b corrects blurring in the X-ray image obtained by the imaging system 20, i.e., performs processing to reduce or remove blurring in the X-ray image. Specifically, the image processing system 10b generates blurring information used to correct blurring in the X-ray image. The image processing system 10b is realized by the same hardware as the image processing system 10a of the first embodiment.

[0090] Next, the functions of the image processing system 10b according to this embodiment will be described. As shown in Fig. 18, the image processing system 10b includes an acquisition unit 11b and a blur information generation unit 12b.

[0091] The acquisition unit 11b is an acquisition means for acquiring inference information, which is at least one of an X-ray image obtained by imaging with the imaging system 20 that performs imaging using X-rays, the position of the object of the imaging in the imaging direction, information related to the X-ray source 21 included in the imaging system 20, and information related to the detector 22 included in the imaging system 20. The information related to the position of the object of the imaging in the imaging direction may be a magnification ratio corresponding to the position. The information related to the X-ray source 21 included in the imaging system 20 may be information indicating the size of the focal point of the X-ray source 21. The information related to the detector 22 included in the imaging system 20 may be at least one of information related to the scintillator included in the radiation detector and information indicating the size of the pixels.

[0092] The acquisition unit 11b, like the acquisition unit 11a of the first embodiment, acquires the same X-ray image and inference information as in the first embodiment. The acquisition unit 11b outputs the acquired X-ray image and inference information to the blur information generation unit 12b.

[0093] The blur information generating unit 12b is a blur information generating means that inputs information based on the X-ray image and inference information acquired by the acquiring unit 11b into an inference model generated by machine learning training, and generates blur information that indicates the response to X-rays in the X-ray image obtained by imaging with the imaging system 20.

[0094] The blur information generated by the image processing system 10b is, for example, a PSF. The PSF is, for example, a two-dimensional spatial intensity distribution. However, the blur information does not need to be a PSF, and may be anything that can be used to correct blurring that occurs in an X-ray image. For example, the blur information may be an OTF (Optical Transfer Function) that expresses the PSF in frequency space, or an MTF (Modulation Transfer Function) that is the absolute value of the OTF.

[0095] The blur information generated by the image processing system 10b is, for example, out-of-focus blur information used to correct out-of-focus blur. The blur information generated by the image processing system 10b may also be blur information used to correct blur due to factors other than out-of-focus blur. For example, the blur information generated by the image processing system 10b may be scintillator blur information used to correct scintillator blur. The blur information generated by the image processing system 10b may also be two types of blur information, for example, the out-of-focus blur information and blur information related to blur due to factors other than out-of-focus blur. The blur information generated by the image processing system 10b may also be blur information used to correct blur including out-of-focus blur and blur due to other factors.

[0096] The blur information generating unit 12b uses an inference model generated by machine learning training to generate blur information. The inference model includes, for example, a neural network. The neural network may be multi-layered. That is, the inference model may be generated by deep learning. The neural network may be a convolutional neural network (CNN). The neural network may be multi-modal. For example, the neural network may input both an image (image information) and a parameter or condition (text information) as shown below. The format of the inference model may be similar to that of an inference model generated by conventional machine learning training.

[0097] For example, as shown in Figure 19, inference model 30b inputs information based on an X-ray image 31 and inference information and outputs blur information 33. In this case, inference model 30b has neurons in the input layer for inputting information based on the X-ray image 31 and inference information. This is similar to inference model 30a described in the first embodiment.

[0098] Furthermore, the inference model 30b is provided with neurons for outputting blur information 33 to the output layer. For example, the information output from the inference model is the intensity value (two-dimensional spatial intensity distribution) at each position, which is the PSF. In this case, the output layer is provided with neurons equal to the number of positions of the PSF, and each neuron outputs a corresponding intensity value. For example, as shown in FIG. 19, the inference model 30a is provided with a channel (1ch) for outputting the PSF.

[0099] The inference model may be anything other than one that outputs blur information, as long as it is used to generate blur information from information based on an X-ray image and information for inference.

[0100] The inference model functions to input information based on X-ray images and inference information, perform calculations according to the input, and output information. The inference model is expected to be used as a program module that is part of artificial intelligence software. The inference model is used, for example, in a computer having a processor and memory, and the computer's processor operates according to instructions from the model stored in the memory. For example, the computer's processor operates according to the instructions to input information to the model, perform calculations according to the model, and output results from the model. Specifically, the computer's processor operates according to the instructions to input information to the input layer of a neural network, perform calculations based on parameters such as learning weighting coefficients in the neural network, and output results from the output layer of the neural network. Note that the inference model may be configured using something other than a neural network.

[0101] The inference model used by the blur information generation unit 12b is generated in advance by machine learning training and stored in the blur information generation unit 12b. Generation of the inference model (machine learning training) will be described. To generate the inference model, a combination of learning images, inference information, and blur information (learning data, teacher data) is prepared. The learning images are images that include blur when captured by an imaging system. Note that the imaging system related to the learning images may be the imaging system 20 that captures images to obtain images to be corrected for blur, or it does not have to be the imaging system 20. The learning images are images of the same size as the X-ray images (X-ray images to be corrected) used to generate blur information by the image processing system 10b.

[0102] The blurred images, inference information, and blur information are prepared in the same manner as when generating the inference model of the first embodiment described above. Note that the blurred images, inference information, and blur information may be prepared by methods other than those described above. The number of combinations of training images, inference information, and blur information prepared is sufficient to properly train the inference model.

[0103] The inference model is trained as follows using the prepared training image, inference information, and blur information. The inference model is trained for each combination of training image, inference information, and blur information. For example, information based on the training image containing blur and the inference information is input to the inference model, and calculations are performed according to the inference model to obtain blur information. The obtained blur information is compared with the training blur information, and the parameters of the inference model are updated by error backpropagation based on a loss resulting from the comparison. In this case, when the training image containing blur is input to the inference model, the inference model is trained so that the training blur information is output from the inference model.

[0104] The inference model may be generated by a method other than the above. The inference model may be generated in the image processing system 10b, or by a system other than the image processing system 10b.

[0105] The blur information generation unit 12b generates blur information, for example, as follows: The blur information generation unit 12b inputs the X-ray image and the information for inference from the acquisition unit 11b. The blur information generation unit 12b inputs information based on the input X-ray image and the information for inference into an inference model to generate blur information. For example, if the inference model outputs blur information, the blur information generation unit 12b acquires the output from the inference model as blur information. Furthermore, the blur information generation unit 12b may generate blur information by a method other than the above, as long as it uses an inference model.

[0106] The blur information generating unit 12b outputs the generated blur information. For example, the blur information generating unit 12b transmits or outputs the generated blur information to a system or module that corrects blur in an X-ray image using the blur information.

[0107] Alternatively, the blur information generation unit 12b may correct the blur of the X-ray image using the generated blur information. The correction of the blur of the X-ray image using the blur information may be performed by a conventional method. For example, the correction unit 13 may correct the blur of the X-ray image based on the PSF, which is the blur information, using a Wiener filter or the Richardson-Lucy algorithm. Alternatively, the correction may be performed using an inference model generated by machine learning training (e.g., deep learning). In this case, the blur information generation unit 12b outputs the generated blur-corrected X-ray image. The blur-corrected X-ray image may be output in the same manner as a conventional method depending on the purpose of use of the X-ray image. The functions of the image processing system 10b according to this embodiment have been described above.

[0108] Next, an image processing method, which is a process executed by the image processing system 10b according to this embodiment (an operating method performed by the image processing system 10b), will be described using the flowchart of Fig. 20. In this process, the acquisition unit 11b acquires an X-ray image and inference information obtained by imaging by the imaging system 20 (S11, acquisition step). The inference information is at least one of the position of the object in the imaging direction, information related to the X-ray source 21 included in the imaging system 20, and information related to the radiation detector included in the imaging system 20.

[0109] Next, the blur information generation unit 12b inputs information based on the X-ray image and the information for inference into the inference model to generate blur information (S12, blur information generation step). The generated blur information is output from the blur information generation unit 12b to a predetermined output destination (S03). The blur information is used to correct blur in the X-ray image. Alternatively, the blur information generation unit 12b may use the blur information to correct blur in the X-ray image. In this case, the blur-corrected X-ray image is output from the blur information generation unit 12b to a predetermined output destination. The above is the image processing method, which is processing executed by the image processing system 10b according to this embodiment.

[0110] In this embodiment, information based on the X-ray image and the inference information is input to an inference model generated by machine learning training, and blur information is generated for the X-ray image obtained by imaging with the imaging system 20. The generated blur information can be used to correct the blur in the X-ray image obtained by imaging with the imaging system 20. Therefore, according to this embodiment, blur in the X-ray image can be corrected without actually measuring the blur for each imaging system 20.

[0111] Furthermore, similarly to the first embodiment, by using the above-described inference information, blurring of the X-ray image can be corrected more appropriately.

[0112] Next, an image processing program for executing a series of processes by the image processing system 10b described above will be described. As shown in Fig. 21, the image processing program 100b is stored in a program storage area 111b formed on a computer-readable recording medium 110b that is inserted into a computer and accessed, or that is provided in the computer. The recording medium 110b may be a non-transitory recording medium.

[0113] The image processing program 100b includes an acquisition module 101b and a blur information generation module 102b. Functions realized by executing the acquisition module 101b and the blur information generation module 102b are similar to the functions of the acquisition unit 11b and the blur information generation unit 12b of the image processing system 10b described above, respectively.

[0114] The image processing program 100b may be configured such that a part or all of it is transmitted via a transmission medium such as a communication line, and is received and recorded (including installed) by another device. Furthermore, each module of the image processing program 100b may be installed not on one computer but on one of multiple computers. In this case, the above-described series of processes are performed by a computer system consisting of the multiple computers.

[0115] The image processing method, image processing system, image processing program, and recording medium disclosed herein have the following configurations. [1] An acquisition step of acquiring information for inference, which is at least one of a radiation image obtained by imaging with an imaging system that performs imaging using radiation, information related to the position of the object of the imaging in the imaging direction, information related to a radiation source included in the imaging system, and information related to a radiation detector included in the imaging system; a correction step of correcting blur in the radiographic image by inputting information based on the radiographic image and the inference information acquired in the acquisition step into an inference model generated by machine learning training; A method for image processing comprising: [2] an acquisition step of acquiring information for inference, which is at least one of a radiation image obtained by imaging with an imaging system that performs imaging using radiation, information related to the position of the object of the imaging in the imaging direction, information related to a radiation source included in the imaging system, and information related to a radiation detector included in the imaging system; a blur information generating step of inputting information based on the radiation image and the inference information acquired in the acquisition step into an inference model generated by machine learning training, to generate blur information indicating a response to radiation in the radiation image acquired by imaging with the imaging system; A method for image processing comprising: [3] The image processing method according to [1] or [2], wherein the information relating to the position of the object in the imaging direction is a magnification factor according to the position. [4] The image processing method according to any one of [1] to [3], wherein the information relating to the radiation source included in the imaging system is information indicating the size of the focal spot of the radiation source. [5] The image processing method according to any one of [1] to [4], wherein the information relating to the radiation detector included in the imaging system is at least one of information relating to the scintillator included in the radiation detector and information indicating the size of the pixels. [6] An acquisition means for acquiring information for inference, which is at least one of a radiation image obtained by imaging with an imaging system that performs imaging using radiation, a position of an object in the imaging direction, information related to a radiation source included in the imaging system, and information related to a radiation detector included in the imaging system; a correction means for correcting blur in the radiographic image by inputting information based on the radiographic image and the inference information acquired by the acquisition means into an inference model generated by machine learning training; and An image processing system comprising: [7] An acquisition means for acquiring information for inference, which is at least one of a radiation image obtained by imaging with an imaging system that performs imaging using radiation, information related to the position of the object in the imaging direction, information related to a radiation source included in the imaging system, and information related to a radiation detector included in the imaging system; a blur information generating means for inputting information based on the radiation image and the inference information acquired by the acquisition means into an inference model generated by machine learning training, and generating blur information indicating a response to radiation in the radiation image acquired by the imaging system; An image processing system comprising: [8] Computer, an acquisition means for acquiring information for inference, which is at least one of a radiation image obtained by imaging with an imaging system that performs imaging using radiation, a position of an object in the imaging direction, information related to a radiation source included in the imaging system, and information related to a radiation detector included in the imaging system; a correction means for correcting blur in the radiographic image by inputting information based on the radiographic image and the inference information acquired by the acquisition means into an inference model generated by machine learning training; and An image processing program that functions as a [9] Computer, an acquisition means for acquiring information for inference, which is at least one of a radiation image obtained by imaging with an imaging system that performs imaging using radiation, information relating to the position of the object in the imaging direction, information relating to a radiation source included in the imaging system, and information relating to a radiation detector included in the imaging system; a blur information generating means for inputting information based on the radiation image and the inference information acquired by the acquisition means into an inference model generated by machine learning training, and generating blur information indicating a response to radiation in the radiation image acquired by the imaging system; An image processing program that functions as a

[10] A computer-readable recording medium having the image processing program described in [8] or [9] recorded thereon. [Explanation of symbols]

[0116] 10a, 10b...image processing system, 11a, 11b...acquisition unit, 12a...correction unit, 12b...blur information generation unit, 20...imaging system, 21...X-ray source, 22...detector, 100a, 100b...image processing program, 101a, 101b...acquisition module, 102a...correction module, 102b...blur information generation module, 110a, 110b...recording medium, 111a, 111b...program storage area.

Claims

1. an acquisition step of acquiring information for inference, which is at least one of a radiation image obtained by imaging with an imaging system that performs imaging using radiation, information related to the position of the object of the imaging in the imaging direction, information related to a radiation source included in the imaging system, and information related to a radiation detector included in the imaging system; a correction step of correcting blur in the radiographic image by inputting information based on the radiographic image and the inference information acquired in the acquisition step into an inference model generated by machine learning training; A method for image processing comprising:

2. an acquisition step of acquiring information for inference, which is at least one of a radiation image obtained by imaging with an imaging system that performs imaging using radiation, information related to the position of the object in the imaging direction, information related to a radiation source included in the imaging system, and information related to a radiation detector included in the imaging system; a blur information generating step of inputting information based on the radiation image and the inference information acquired in the acquisition step into an inference model generated by machine learning training, to generate blur information indicating a response to radiation in the radiation image acquired by imaging with the imaging system; A method for image processing comprising:

3. 3. The image processing method according to claim 1, wherein the information relating to the position of the object in the imaging direction is a magnification factor according to the position.

4. 3. The image processing method according to claim 1, wherein the information relating to the radiation source included in the imaging system is information indicating the size of the focal spot of the radiation source.

5. 3. The image processing method according to claim 1, wherein the information relating to the radiation detector included in the imaging system is at least one of information relating to a scintillator included in the radiation detector and information indicating a pixel size.

6. an acquisition means for acquiring information for inference, which is at least one of a radiation image obtained by imaging with an imaging system that performs imaging using radiation, information relating to the position of the object in the imaging direction, information relating to a radiation source included in the imaging system, and information relating to a radiation detector included in the imaging system; a correction means for correcting blur in the radiographic image by inputting information based on the radiographic image and the inference information acquired by the acquisition means into an inference model generated by machine learning training; and An image processing system comprising:

7. an acquisition means for acquiring information for inference, which is at least one of a radiation image obtained by imaging with an imaging system that performs imaging using radiation, information relating to the position of the object in the imaging direction, information relating to a radiation source included in the imaging system, and information relating to a radiation detector included in the imaging system; a blur information generating means for inputting information based on the radiation image and the inference information acquired by the acquisition means into an inference model generated by machine learning training, and generating blur information indicating a response to radiation in the radiation image acquired by the imaging system; An image processing system comprising:

8. Computer, an acquisition means for acquiring information for inference, which is at least one of a radiation image obtained by imaging with an imaging system that performs imaging using radiation, information relating to the position of the object in the imaging direction, information relating to a radiation source included in the imaging system, and information relating to a radiation detector included in the imaging system; a correction means for correcting blur in the radiographic image by inputting information based on the radiographic image and the inference information acquired by the acquisition means into an inference model generated by machine learning training; and An image processing program that functions as a

9. Computer, an acquisition means for acquiring information for inference, which is at least one of a radiation image obtained by imaging with an imaging system that performs imaging using radiation, information relating to the position of the object in the imaging direction, information relating to a radiation source included in the imaging system, and information relating to a radiation detector included in the imaging system; a blur information generating means for inputting information based on the radiation image and the inference information acquired by the acquisition means into an inference model generated by machine learning training, and generating blur information indicating a response to radiation in the radiation image acquired by the imaging system; An image processing program that functions as a

10. 10. A computer-readable recording medium on which the image processing program according to claim 8 or 9 is recorded.

Images