Image processing apparatus, radiography system, image processing method, and program

Abstract

To appropriately perform image processing even when a learned model which has learned learning data obtained by an imaging apparatus having characteristics different from those of an imaging apparatus related to a processing target is used.SOLUTION: Provided is an image processing apparatus which includes: an acquisition unit configured to acquire a first radiographic image obtained using a first imaging apparatus; and an image processing unit configured to perform image processing on the first radiographic image using a learned model obtained using learning data based on a second radiographic image obtained using a second imaging apparatus having different characteristics. A characteristic of at least one of the first radiographic image and the second radiographic image is adjusted to ensure that their characteristics match. The learning data includes the second radiographic image or the second radiographic image in which the characteristic is adjusted. When the second radiographic image having the characteristics adjusted is not included in the learning data, image processing of the first radiographic image is performed by inputting the first radiographic image having the characteristics adjusted to the learned model.SELECTED DRAWING: Figure 4C

Description

[Technical field]

[0001] The present disclosure relates to an image processing device, a radiation imaging system, an image processing method, and a program. [Background technology]

[0002] In recent years, radiation imaging devices equipped with a detection unit for detecting radiation such as X-rays have been widely used in fields such as industry and medicine. In particular, digital radiography (DR) devices that obtain radiation images using semiconductor sensors have become widespread.

[0003] In such digital radiography devices, various image processing is generally performed to improve the image quality of the captured image, and one of these is noise reduction processing that improves the graininess of the captured image and improves the visibility of the diagnostic area. In particular, noise reduction processing that applies machine learning technology such as deep learning has the potential to significantly improve graininess. For example, Patent Document 1 proposes a technology that obtains a good noise reduction effect by learning noise characteristics according to the characteristics of the imaging device. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent Publication No. 2022-39989 Summary of the Invention [Problem to be solved by the invention]

[0005] However, the above-mentioned conventional techniques may have the following problems: According to Patent Document 1, in order to learn noise characteristics according to the characteristics of an imaging device and obtain a good noise reduction effect, it is necessary to prepare appropriate learning data obtained by an imaging device having characteristics similar to those of the processing target.

[0006] Here, imaging devices that require an invasive imaging process such as digital radiography are characterized by the very high cost of collecting a data set, given the risk of radiation exposure. On the other hand, imaging devices are constantly being improved due to recent technological developments, and various characteristics of imaging devices such as resolution, noise characteristics, and sensitivity continue to be improved.

[0007] Therefore, it is often difficult to prepare learning data obtained from an imaging device with characteristics similar to those of the processing target, and in such cases, it may not be possible to achieve the intended noise reduction performance. Note that although issues related to noise reduction processing have been described here, similar issues can also arise in other image processing, such as super-resolution processing, gradation processing, and grid stripe reduction processing.

[0008] An embodiment of the present disclosure aims to provide an image processing device that can perform appropriate image processing even when using a trained model that has been trained using learning data obtained by an imaging device with characteristics different from those of the imaging device to be processed. [Means for solving the problem]

[0009] An image processing device according to an embodiment of the present disclosure includes an acquisition unit that acquires a first radiographic image of an object to be inspected obtained using a first imaging device, and an image processing unit that performs image processing of the first radiographic image using a trained model obtained using training data based on a second radiographic image obtained using a second imaging device having characteristics different from the characteristics of the first imaging device, wherein image characteristics of at least one of the first radiographic image and the second radiographic image are adjusted so that characteristics of the first radiographic image and characteristics of the second radiographic image match, and the training data includes the second radiographic image or a second radiographic image with the image characteristics adjusted, and when the training data does not include a second radiographic image with the image characteristics adjusted, the image processing unit performs image processing of the first radiographic image by inputting the first radiographic image with the image characteristics adjusted to the trained model. Effect of the Invention

[0010] According to one embodiment of the present disclosure, image processing can be performed appropriately even when using a trained model that has been trained using learning data obtained by an imaging device with characteristics different from those of the imaging device to be processed. [Brief description of the drawings]

[0011] [Figure 1A] 1 shows a schematic configuration of a radiation imaging system according to a first embodiment. [Figure 1B] 2 shows a schematic configuration of a detection unit according to the first embodiment. [Figure 2A] 2 illustrates a schematic configuration of a control unit according to the first embodiment. [Figure 2B] 2 illustrates a schematic configuration of a noise reduction processing unit according to a first embodiment. [Figure 3A] 1 illustrates an example of a schematic configuration of a trained model according to a first embodiment. [Figure 3B] 1 illustrates an example of a schematic configuration of a CNN according to a first embodiment. [Figure 3C] FIG. 4 is a diagram for explaining an operation example of a learning process according to the first embodiment. [Figure 4A] 4 shows an example of MTF characteristics of the imaging apparatus according to the first embodiment. [Figure 4B] 13 illustrates an example of a process flow of an inference process according to the first embodiment. [Figure 4C] 3A to 3C are schematic diagrams of images generated in the flow of the inference process and the learning process according to the first embodiment. [Figure 4D] 1 illustrates an example of a processing flow of a learning process according to the first embodiment. [Figure 5A] 13 illustrates an example of a process flow of an inference process according to the second embodiment. [Figure 5B] 11 is a schematic diagram of an image generated in the flow of the inference process and the learning process according to the second embodiment. FIG. [Figure 6A] 13 illustrates an example of a process flow of an inference process according to the third embodiment. [Figure 6B] 13 is a schematic diagram of an image generated in the flow of the inference process and the learning process according to the third embodiment. FIG. [Figure 6C] 13 illustrates an example of a processing flow of a learning process according to the third embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0012] Hereinafter, exemplary embodiments for carrying out the present disclosure will be described in detail with reference to the drawings. However, the dimensions, materials, shapes, and relative positions of components described in the following embodiments are arbitrary and can be changed according to the configuration of the device to which the present disclosure is applied or various conditions. In addition, the same reference numerals are used in the drawings to indicate elements that are the same or functionally similar.

[0013] In the following, a radiography system using X-rays as an example of radiation will be described. However, the radiation may be X-rays or other radiation. In the following examples, the term radiation may include, for example, electromagnetic radiation such as X-rays and gamma rays, and particle radiation such as alpha rays, beta rays, particle beams, proton beams, heavy ion beams, and meson beams.

[0014] In the following, the machine learning model refers to a learning model based on a machine learning algorithm. Specific examples of machine learning algorithms include nearest neighbor methods, naive Bayes methods, decision trees, and support vector machines. Neural networks and deep learning may also be used. Appropriately, any of the above algorithms that can be used can be used in the following embodiments and modifications. Furthermore, the learning data refers to a data set used for learning the machine learning model, and is composed of a pair of input data input to the machine learning model and teacher data that is the correct answer for the output result of the machine learning model.

[0015] A trained model refers to a model that has been trained in advance using appropriate training data for a machine learning model that follows any machine learning algorithm, such as deep learning. However, although a trained model has been obtained by training in advance using appropriate training data, this does not mean that no further training is performed, and additional training can be performed. Additional training can also be performed after the device is installed at the site of use.

[0016] Example 1 (Configuration of Radiography System) An image processing device, a radiography system, and an image processing method according to a first embodiment of the present disclosure will be described below with reference to Fig. 1A and Fig. 1B. Fig. 1A shows a schematic configuration of a radiography system 1 according to this embodiment. In the following description, an object O to be inspected is a human body, but the object O to be inspected photographed by the radiography system according to the present disclosure is not limited to a human body, and may be other animals, plants, objects to be subjected to non-destructive inspection, etc.

[0017] The radiation imaging system 1 according to this embodiment is provided with a detection unit 10, a control unit 20, a radiation generator 30, an input unit 40, and a display unit 50. The radiation imaging system 1 may also include an external storage device 70, such as a server, connected to the control unit 20 via a network 60, such as the Internet or an intranet.

[0018] The radiation generator 30 includes a radiation generating source such as an X-ray tube and is capable of emitting radiation. The detection unit 10 detects the radiation emitted by the radiation generator 30 and is capable of generating a radiation image corresponding to the detected radiation. Thus, the detection unit 10 functions as an example of a radiography device capable of generating a radiation image of the object O to be inspected by detecting the radiation emitted by the radiation generator 30 and passing through the object O to be inspected.

[0019] Here, FIG. 1B shows an example of a schematic configuration of the detection unit 10 according to this embodiment. The detection unit 10 is provided with a phosphor 11 and an imaging sensor 12. The phosphor 11 converts radiation incident on the detection unit 10 into light of a wavelength that can be detected by the imaging sensor 12. The phosphor 11 is, for example, CsI or GOS (Gd 2 O 2 The imaging sensor 12 may include a photoelectric conversion element made of, for example, a-Si or crystalline Si, and can detect light corresponding to the radiation converted by the phosphor 11 and output a signal corresponding to the detected light. The detection unit 10 can generate a radiation image by performing A / D conversion or the like on the signal output by the imaging sensor 12. Although not shown in FIG. 1B, the detection unit 10 may include a calculation unit, an A / D conversion unit, and the like. In addition, a grid may be provided between the detection unit 10 and the object O to be inspected, so that scattered rays that are generated when radiation passes through the object O and reach the detection unit 10 can be reduced.

[0020] As described above, in this embodiment, an indirect conversion type detector that converts radiation into visible light using a scintillator or the like and then converts the visible light into an electrical signal using a photoelectric conversion element is used as the detection unit 10. Alternatively, a direct conversion type detector that directly converts incident radiation into an electrical signal may be used as the detection unit 10.

[0021] The control unit 20 is connected to the detection unit 10, the radiation generator 30, the input unit 40, and the display unit 50. The control unit 20 can acquire a radiation image output from the detection unit 10 and perform image processing on the radiation image, and control the driving of the detection unit 10 and the radiation generator 30. This allows the control unit 20 to generate radiation under predetermined imaging conditions at appropriate timing. The control unit 20 can also function as an example of an image processing device. Furthermore, the control unit 20 may be connected to an external storage device 70 via an arbitrary network 60 such as the Internet or an intranet, and may acquire a radiation image or the like from the external storage device 70. Furthermore, the control unit 20 may be connected to other detection units, radiation generators, or the like via the network 60. The control unit 20 may be connected to the external storage device 70 or the like via a wired or wireless connection.

[0022] The input unit 40 includes input devices such as a mouse, a keyboard, a trackball, a touch panel, etc., and can be operated by an operator to input instructions to the control unit 20. The display unit 50 includes, for example, any monitor, and can display information and images output from the control unit 20, information input by the input unit 40, etc.

[0023] In this embodiment, the control unit 20, input unit 40, display unit 50, etc. are configured as separate devices, but they may be configured as an integrated unit. For example, the input unit 40 and display unit 50 may be configured as a touch panel display. In this embodiment, the control unit 20 configures an image processing device, but it is only necessary for the image processing device to be able to acquire a radiation image and perform image processing on the radiation image, and it is not necessary for the image processing device to control the driving of the detection unit 10 or the radiation generator 30.

[0024] The control unit 20, the detection unit 10, the radiation generator 30, etc. may be connected to each other by wire or wirelessly. Furthermore, the external storage device 70 may constitute an image system such as a PACS in a hospital, or may be a server outside the hospital.

[0025] (Control Unit Configuration) Next, a more specific configuration of the control unit 20 will be described with reference to Fig. 2A and Fig. 2B. Fig. 2A shows a schematic configuration of the control unit 20 according to this embodiment, and Fig. 2B shows a schematic configuration of the noise reduction processing unit 26. The control unit 20 is provided with an acquisition unit 21, an image processing unit 22, a display control unit 23, a drive control unit 24, and a storage unit 25.

[0026] The acquiring unit 21 can acquire the radiation image output by the detecting unit 10, various information input by the input unit 40, etc. The acquiring unit 21 can also acquire the radiation image, patient information, etc. from the external storage device 70, etc.

[0027] The image processing unit 22 is provided with a noise reduction processing unit 26 and a diagnostic image processing unit 27, and can perform image processing according to the present disclosure on the radiological image acquired by the acquisition unit 21. In this embodiment, noise reduction processing will be described as an example of image processing performed by the image processing unit 22.

[0028] 2B, the noise reduction processing unit 26 is provided with a learning processing unit 261 and an inference processing unit 262. In addition to the configuration of the inference processing unit 262, the learning processing unit 261 is further provided with a learning data selection unit 263, a learning data generation unit 264, a parameter update unit 265, a frequency adjustment unit 266, and a conversion processing unit 267. With this configuration, the noise reduction processing unit 26 can learn a machine learning model for performing noise reduction processing, and apply noise reduction processing suitable for radiation images using the machine learning model.

[0029] The inference processing unit 262 uses the acquired radiation image as input data for a trained model trained using the learning processing unit 261, thereby inferring an image with reduced noise.

[0030] Here, a description will be given of the configuration of the learning processing unit 261. The learning processing unit 261 performs a learning process applied when learning a machine learning model, and in addition to the configuration of the inference processing unit 262, further includes a learning data selection unit 263, a learning data generation unit 264, a parameter update unit 265, a frequency adjustment unit 266, and a conversion processing unit 267.

[0031] The learning data selection unit 263 selects images (data) to be used as learning data from the images input to the learning processing unit 261 when performing the learning process. For example, the learning data selection unit 263 can select a radiation image obtained using a specific imaging device from among the images input to the learning processing unit 261. Note that the selection criteria are not limited to a specific imaging device, and may be, for example, the model of the imaging device, imaging conditions, etc. Here, the imaging conditions may include the pixel size and modulation transfer function (MTF) characteristics of the imaging device, image size, etc.

[0032] The learning data generation unit 264 creates learning data using the data selected by the learning data selection unit 263. In this embodiment, an example of a configuration is shown in which an image to which artificial noise has been added is used as input data and an image to which no artificial noise has been added is used as teacher data, as a set of learning data for learning noise reduction processing. The learning data generation unit 264 performs processing to create a set of learning data by adding artificial noise created by simulating the characteristics of a radiation image to the selected image. Here, the noise to be added by the learning data generation unit 264 can be calculated by the learning data selection unit 263.

[0033] The parameter update unit 265 performs processing to update parameters of the machine learning model held by the inference processing unit 262 based on the inference result by the inference processing unit 262 and the correct answer data. The frequency adjustment unit 266 converts (adjusts) various frequency characteristics such as sharpness of the image. The conversion processing unit 267 converts (adjusts) the image size of the image. Details of these conversion processes will be described later.

[0034] Here, the learning processing unit 261 does not need to be included in the control unit 20. For example, the configuration of the learning processing unit 261 may be configured on hardware other than the control unit 20 such as a server, and a trained model may be created by performing learning in advance using appropriate training data. Note that even when the configuration of the learning processing unit 261 is configured on hardware other than the control unit 20 such as a server, the inference processing unit 262, the frequency adjustment unit 266, and the conversion processing unit 267 may be provided in the noise reduction processing unit 26 separately from the configuration of the learning processing unit 261. In this case, the control unit 20 may access the separate hardware by the inference processing unit 262 and perform only processing using the trained model. Also, a trained model created in advance may be provided in the noise reduction processing unit 26, and the trained model provided with the inference processing unit 262 may be used. Alternatively, the learning processing unit 261 may be included in the control unit 20, allowing additional learning to be performed using training data acquired after installation.

[0035] The diagnostic image processor 27 can perform diagnostic image processing for converting the image that has been subjected to noise reduction processing by the noise reduction processor 26 into an image suitable for diagnosis. For example, the diagnostic image processor 27 may perform gradation processing, emphasis processing, grid stripe reduction processing, etc. on the image that has been subjected to noise reduction processing according to a region of interest (ROI) set in the radiation image. For example, the gradation processing may be performed so as to widen the gradation of the region of interest, and the emphasis processing may be performed so as to emphasize the region of interest. Here, the region of interest may be set according to an instruction from an operator, or may be set based on the imaging site, disease name information, finding information, etc.

[0036] The display control unit 23 can control the display of the display unit 50, and can cause the display unit 50 to display, for example, radiation images before and after image processing by the image processing unit 22, patient information, etc. The drive control unit 24 can control the drive of the detection unit 10, the radiation generator 30, etc. Therefore, the control unit 20 can control the capture of radiation images by controlling the drive of the detection unit 10 and the radiation generator 30 by the drive control unit 24.

[0037] The storage unit 25 can store an operating system (OS), device drivers for peripheral devices, and programs for implementing various application software including programs for performing processes described below. The storage unit 25 can also store information acquired by the acquisition unit 21, radiographic images processed by the image processing unit 22, and the like. For example, the storage unit 25 can store radiographic images acquired by the acquisition unit 21, and radiographic images that have been subjected to noise reduction processing described below.

[0038] The control unit 20 can be configured using a general computer including a processor, a memory, etc., but may be configured as a dedicated computer for the radiation imaging system 1. Here, the control unit 20 functions as an example of an image processing device according to this embodiment, but the image processing device according to this embodiment may be a separate (external) computer communicatively connected to the control unit 20. The control unit 20 and the image processing device may be, for example, a personal computer, and a desktop PC, a notebook PC, or a tablet PC (portable information terminal) may be used. The processor may be a CPU (Central Processing Unit). The processor may be, for example, an MPU (Micro Processing Unit), a GPU (Graphical Processing Unit), or an FPGA (Field-Programmable Gate Array), etc.

[0039] Each function of the control unit 20 may be realized by a processor such as a CPU or MPU executing a software module stored in the storage unit 25. The processor may be, for example, a GPU or FPGA. Each function may be configured by a circuit that performs a specific function, such as an ASIC. For example, the image processing unit 22 may be realized by dedicated hardware such as an ASIC, and the display control unit 23 may be realized by a dedicated processor such as a GPU that is different from the CPU. The storage unit 25 may be configured by any storage medium, such as an optical disk such as a hard disk, or a memory.

[0040] (Machine learning model configuration) Next, an example of a machine learning model constituting the trained model according to this embodiment will be described with reference to Figures 3A to 3C. An example of the machine learning model used by the inference processing unit 262 according to this embodiment is a multi-layer neural network.

[0041] Fig. 3A shows a schematic configuration example of a neural network model according to this embodiment. The neural network model 33 shown in Fig. 3A is designed to output inference data 32 in which noise has been reduced according to a previously learned tendency for input data 31. The output radiographic image in which noise has been reduced is based on the learning content in the machine learning process, and the neural network according to this embodiment has learned feature amounts for discriminating between signals and noise contained in the radiographic image.

[0042] At least a part of the multi-layered neural network may be, for example, a convolutional neural network (hereinafter, referred to as a CNN). Also, at least a part of the multi-layered neural network may be based on a technology related to an autoencoder.

[0043] Here, a case where CNN is used as a machine learning model for noise reduction processing of a radiological image will be described. Fig. 3B shows an example of a schematic configuration 34 of a CNN constituting a neural network model according to this embodiment. In the example of the trained model according to this embodiment, when input data 31, which is a radiological image, is input, inference data 32 in which noise has been reduced can be output.

[0044] The CNN shown in FIG. 3B is composed of a plurality of layers that process and output a group of input values. The types of layers included in the CNN configuration 34 include a convolution layer, a downsampling layer, an upsampling layer, and a merge layer. The CNN configuration 34 may also include an addition layer 35. In this case, the CNN configuration 34 may include a shortcut that adds input data using the addition layer 35 before output. This allows the CNN to be configured to learn the difference between input data and output data, and to suitably handle a system that targets noise.

[0045] The convolution layer is a layer that performs convolution processing on a group of input values ​​according to parameters such as the set filter kernel size, the number of filters, the stride value, the dilation value, etc. The number of dimensions of the filter kernel size may also be changed according to the number of dimensions of the input image.

[0046] The downsampling layer is a layer that performs processing to reduce the number of output value groups to be less than the number of input value groups by thinning out or combining input value groups. Specifically, such processing includes, for example, Max Pooling processing.

[0047] An upsampling layer is a layer that performs a process to make the number of output values ​​larger than the number of input values ​​by duplicating the input values ​​or adding values ​​interpolated from the input values, etc. Specifically, such a process includes, for example, upsampling by deconvolution.

[0048] A synthesis layer is a layer that inputs a group of values, such as a group of output values ​​of a layer or a group of pixel values ​​that make up an image, from multiple sources and performs processing to synthesize them by concatenating or adding them.

[0049] Note that different parameter settings for the layers and nodes that make up the neural network may affect the degree to which the trends trained from the learning data can be reproduced during inference. In other words, in many cases, appropriate parameters differ depending on the implementation form, so they can be changed to preferred values ​​as necessary.

[0050] In addition to changing the parameters as described above, CNNs may be able to obtain better characteristics by changing their configurations, such as outputting radiological images with reduced noise with higher accuracy, shortening processing time, or shortening the time required to train a machine learning model.

[0051] The CNN configuration 34 used in this embodiment is a U-net type machine learning model having an encoder function consisting of multiple layers including multiple downsampling layers and a decoder function consisting of multiple layers including multiple upsampling layers. In the U-net type machine learning model, the position information (spatial information) made ambiguous in the multiple layers configured as the encoder can be used in the same dimensional layers (layers corresponding to each other) in the multiple layers configured as the decoder (for example, by using skip connections).

[0052] Although not shown, as an example of a modification of the CNN configuration, for example, a layer of an activation function (for example, ReLu: Rectifier Linear Unit) may be incorporated before and after the convolution layer.

[0053] Through these steps of CNN, noise features can be extracted from the input radiological image.

[0054] 3C, the parameter update unit 265 can calculate a loss function from inference data 32 obtained by applying the neural network model 33 of the inference processing unit 262 to input data 31 in the training data, and supervised data 36 in the training data. The parameter update unit 265 can perform a process of updating the parameters of the neural network model 33 based on the calculated loss function. Here, the loss function indicates the error between the inference data 32 and the supervised data 36.

[0055] Furthermore, the parameter update unit 265 can update the filter coefficients of the convolution layer, for example, by using the backpropagation method so as to reduce the error between the inference data 32 represented by the loss function and the ground truth data 36. The backpropagation method is a method of adjusting parameters between each node of the neural network so as to reduce the above error. Note that the learning may be performed using a method (dropout) of randomly inactivating units (each neuron or each node) that constitute the CNN. The detailed operation of the learning processing unit 261 will be described later.

[0056] Furthermore, the trained model used by the inference processing unit 262 may be one generated using transfer learning. In this case, for example, transfer learning may be performed on a machine learning model trained on a radiographic image of an object O to be inspected that is different in type, etc., to generate a trained model to be used in noise reduction processing. By performing such transfer learning, a trained model can be efficiently generated even for an object O to be inspected that is difficult to obtain a large amount of training data. The object O to be inspected that is different in type, etc., referred to here may be, for example, an animal, a plant, or an object to be subjected to non-destructive inspection.

[0057] Here, the GPU can perform efficient calculations by processing more data in parallel. Therefore, when performing learning multiple times using a learning model using CNN as described above, it is effective to perform processing using the GPU. Therefore, the learning processing unit 261 according to this embodiment uses a GPU in addition to a CPU. Specifically, when executing a learning program including a learning model, the CPU and the GPU work together to perform calculations to perform learning. Note that in the learning process, calculations may be performed only by the CPU or the GPU. Also, each process of the inference processing unit 262 may be realized using a GPU like the learning processing unit 261.

[0058] The configuration of the machine learning model has been explained above, but it is not limited to the models using CNN as shown above, and any model similar to machine learning that uses a model that can extract (express) the features of training data such as images through learning itself can be used.

[0059] Here, the learning processing unit 261 according to this embodiment can use any set of learning data for learning the noise reduction process. For example, learning can be performed using an image with artificial noise added as input data and an image without artificial noise added as teacher data.

[0060] (Image processing unit) Next, the detailed operation of the image processing unit 22 in the radiation imaging system 1 will be described with reference to Figs. 4A to 4D. Here, the radiation imaging system 1 has a semiconductor detector with a pixel size P1 [µm] in the detection unit 10. Hereinafter, the detection unit 10 having the semiconductor detector with the pixel size P1 will be referred to as the first imaging device. By performing imaging using the first imaging device, a first radiation image corresponding to the pixel size P1 of the detector can be obtained. Here, consider a case where the second radiation image used as learning data for the noise reduction process performed by the noise reduction processing unit 26 has different characteristics from the first radiation image obtained by the first imaging device. In this embodiment, for example, it is assumed that the second radiation image is obtained by a second imaging device equipped with a detection unit having a semiconductor detector with a larger pixel size P2 [µm] (P1 <P2のケース)。

[0061] In the following, pixel size refers to the size of each pixel in the detector of the imaging device. In addition, in a radiation image captured using the imaging device, the signal of each pixel constituting the image is obtained by sampling the subject information according to the pixel size of the detector, and it can be said that the image has pixel size information. In light of this, in the following, the pixel size information of each pixel in the radiation image is referred to as the pixel size of the image. Here, as exemplified below, the pixel size of the radiation image can be converted in size using appropriate image processing.

[0062] Now, when a first radiographic image is acquired using a first imaging device with a pixel size of P1 [μm], each pixel constituting the first radiographic image has subject information in a rectangular area with one side of P1 [μm], and the pixel size of the image is P1. When the first imaging device has an imaging area of ​​L [mm] square, the number of pixels on one side of the first radiographic image wp1 = L / P1 x 1000 [pixels]. Similarly, when a second imaging device with a pixel size of P2 [μm] also has an imaging area of ​​L [mm] square, when a second radiographic image is acquired, the number of pixels on one side wp2 = L / P2 x 1000 [pixels]. Here, even if the image size of the radiographic image is converted by enlarging or reducing, the imaging area L does not change. Therefore, for example, if the image size of the second radiographic image is converted so that it becomes (P2 / P1) times larger, the number of pixels on one side of the converted radiographic image will be L / P1×1000=wp1, and it can be seen that the image can be converted into a radiographic image with a pixel size equivalent to P1. As described above, the pixel size of the radiographic image can be converted by changing the image size of the radiographic image.

[0063] FIG. 4A shows an example of modulation transfer function (MTF) characteristics of the first imaging device and the second imaging device. More specifically, FIG. 4A shows MTF1 (first MTF 411) as an example of the MTF characteristics of the first imaging device, and MTF2 (second MTF 412) as an example of the MTF characteristics of the second imaging device. At this time, the Nyquist frequency Nq1 of the first radiographic image is higher than the Nyquist frequency Nq2 of the second radiographic image. Therefore, when the second radiographic image is used as learning data, the machine learning model cannot learn the image and noise characteristics between the spatial frequencies Nq2 and Nq1 in the image (spatial frequency band 413). Therefore, when such a trained model is used, there is a problem that it is not possible to appropriately perform noise reduction processing on the first radiographic image obtained using the first imaging device.

[0064] In contrast, in the image processing unit 22 according to the present embodiment, by matching the characteristics of the input data used for the inference process with the characteristics of the learning data, appropriate noise reduction processing can be performed using the learned model. First, the inference process in the image processing unit 22 will be described with reference to FIGS. 4B and 4C. FIG. 4B is a processing flow of the inference process in the image processing unit 22. FIG. 4C is a schematic diagram of the images generated by the inference process flow and the learning process flow of the image processing unit 22.

[0065] In step S401, the acquisition unit 21 acquires a first radiation image 41 obtained by a first imaging device having a semiconductor detector with a pixel size P1 [μm] and a modulation transfer function of MTF1. The acquired first radiation image 41 is sent to the image processing unit 22. Note that the acquisition unit 21 may acquire the first radiation image from the first imaging device or from an external storage device 70 or the like.

[0066] In step S402, the first radiation image 41 is subjected to image size conversion processing by the conversion processing unit 267 and output as a fourth radiation image 44. The image size conversion processing is to convert the image size of the first radiation image 41 so that the pixel size P1 [μm] of the first radiation image 41 corresponds to the pixel size P2 [μm] of the second radiation image 42 obtained using the second radiation imaging device used as learning data. As a result, assuming the width and height of the first radiation image 41 are width w1 [pixel] and height h1 [pixel], with the magnification factor A = P1 / P2, the width and height of the fourth radiation image 44 are width w1 × A [pixel] and height h1 × A [pixel]. Since P1 < P2 in the image size conversion processing here, A is 1 or less, and the image is reduced. For the pixel value interpolation method associated with the image size conversion, a known interpolation filter such as bilinear interpolation or Lanczos3 interpolation may be used. The spatial frequency characteristic MTF4 of the fourth radiation image 44 is the result of superimposing the influence of the interpolation filter related to the image size conversion processing on the spatial frequency characteristic MTF1.

[0067] In step S403, the inference processing unit 262 performs preprocessing for performing appropriate inference processing on the fourth radiation image 44 obtained in step S402. The method of preprocessing is not limited, but for example, by performing square root transformation, logarithmic transformation, or the like, quantum noise of radiation following a Poisson distribution can be made approximately constant regardless of the intensity of the input radiation. In addition, as preprocessing, a process of converting so that additive noise can be handled, or a process of setting the average value to 0 can be used. Note that the preprocessing is not limited to these, and any appropriate preprocessing can be performed, such as a process of normalizing data with 0-1, or a process of standardizing data so that the average value is 0 and the standard deviation is 1.

[0068] In step S404, the inference processing unit 262 performs inference processing on the result of the preprocessing in step S403 using the trained model 46. More specifically, the inference processing unit 262 inputs the result of the preprocessing to the trained model 46, thereby acquiring the inference processing result output from the trained model 46. A learning method for the trained model 46 will be described later. Note that the trained model 46 may be a machine learning model that has been trained in advance, and it is not necessary for learning to be performed for each series of image processing.

[0069] In step S405, the inference processing unit 262 performs post-processing on the inference processing result of step S404. The post-processing is to perform reverse processing of the conversions such as various normalizations and smoothings performed in the pre-processing in step S403. By performing the post-processing, the inference processing unit 262 can obtain the third radiological image 43 that has been subjected to noise reduction processing.

[0070] In step S406, the diagnostic image processor 27 performs the various diagnostic image processing described above on the third radiographic image 43 obtained in step S405, and outputs an image suitable for diagnosis. The diagnostic image processing in this step is processing suited to the characteristics of the third radiographic image 43, including the pixel size P2 [μm] and the spatial frequency characteristic MTF4.

[0071] By the above-mentioned processing of steps S401 to S406, even if the characteristics of the learning data used in creating the trained model for noise reduction processing differ from the characteristics of the radiation image used in the actual imaging work, it is possible to generate a radiation image with suitably reduced noise. By using such a radiation image, it is possible to support more appropriate image analysis and to support appropriate image diagnosis by a doctor.

[0072] Next, a detailed description will be given of the process of the learning processing unit 261. FIG.

[0073] In step S411, the learning data selection unit 263 selects learning data. As the learning data, an arbitrary image is selected from the image group of the second radiographic images 42 acquired by the second imaging device. The second radiographic images 42 used as the learning data may be acquired using the second imaging device. The acquisition unit 21 acquires appropriate radiographic images acquired using the second imaging device as the learning data. The acquisition unit 21 may acquire radiographic images stored in the storage unit 25, or may acquire radiographic images from the external storage device 70, etc.

[0074] In step S412, the frequency adjustment unit 266 performs frequency adjustment processing on the second radiographic image 42 to generate a fifth radiographic image 45. This is a correction processing for making the spatial frequency characteristics of the fifth radiographic image 45 used as learning data closer to the spatial frequency characteristics MTF4 of the fourth radiographic image 44 used during inference described above. The frequency adjustment unit 266 can apply, for example, a compensation filter to the second radiographic image 42 so that the spatial frequency characteristics MTF2 of the second radiographic image 42 are approximately equal to the spatial frequency characteristics MTF4. Note that if frequency adjustment is not necessary, step S412 may be skipped.

[0075] In step S413, the learning data generation unit 264 adds artificial noise to the fifth radiographic image 45. The artificial noise added here is in accordance with the characteristics of noise to be reduced in the noise reduction process performed at the time of inference. Here, the artificial noise added is noise assumed to be present in the fourth radiographic image 44. For example, this noise is noise that takes into account white noise corresponding to the system noise of the fourth radiographic image 44, quantum noise affected by the spatial frequency characteristic MTF4, and also a change in frequency characteristics due to the interpolation filter used in the image size conversion process performed by the conversion processing unit 267.

[0076] The artificial noise (addNoise) can be calculated as follows. First, the artificial system noise sNoise that simulates the system noise is calculated with a standard deviation of σ s Next, we define an artificial quantum noise qNoise with variance σ q 2 The noise is defined as noise that follows a Poisson distribution and has an NPS (Noise Power Spectrum) according to the MTF of the phosphor mounted on the detection unit 10. For example, when the spatial frequency characteristic MTF4 in the fourth radiation image 44 is approximated by a two-dimensional filter, fMTF, the artificial quantum noise qNoise can be created by convolving fMTF with a noise image that follows a Poisson distribution. Note that since the Poisson distribution can be approximated to a normal distribution when the variance is sufficiently large, the artificial quantum noise qNoise may be treated as noise that follows a normal distribution. In this case, the artificial noise (addNoise) can be expressed as addNoise=C×(sNoise+qNoise). C is an arbitrary coefficient, and usually C=1 is preferable, but if it is desired to change the noise reduction effect, it is also possible to adjust the amount of noise added by changing C.

[0077] It is experimentally known that the distribution of noise occurring in a radiological image approximates a normal distribution with a mean and median of approximately 0, and the artificial noise obtained as described above can also be made to have a distribution with a mean and median of approximately 0. Note that the artificial system noise sNoise may be obtained by taking into consideration the system noise in the first radiological image 41 and the influence of the interpolation filter used in the image size conversion process in step S402.

[0078] In step S413, the learning data generation unit 264 can use the fifth radiographic image 45 to which artificial noise has been added as input data for the learning data. On the other hand, the learning data generation unit 264 can use the fifth radiographic image 45 to which no artificial noise has been added as correct answer data (teacher data) for the learning data. The learning processing unit 261 can generate a trained model that estimates an image with reduced noise from an image containing noise by training a machine learning model using a set of these data as learning data.

[0079] In step S414, the inference processing unit 262 performs appropriate pre-processing on the set of learning data obtained in step S413. This pre-processing is the same as the processing in step S403 described above.

[0080] In step S415, the inference processing unit 262 performs inference processing using a machine learning model on an image obtained by adding artificial noise to the fifth radiographic image 45, which is the input data, among the learning data, and outputs inference data to which noise reduction processing has been applied.

[0081] In step S416, the parameter update unit 265 compares the inference data with the teacher data, and calculates a loss function that quantifies the error between them. The parameter update unit 265 also updates the parameters of the machine learning model (e.g., filter coefficients of a convolutional layer in a CNN) based on the loss function. Examples of the loss function include the mean absolute error (L1loss) and the mean square error (L2loss).

[0082] In step S417, the learning processing unit 261 judges whether learning is to be terminated, and terminates the flow if it is judged that learning is to be terminated. On the other hand, if the learning processing unit 261 judges that learning is not to be terminated, the flow returns to step S411, and the flow of steps S411 to S416 is repeated using other data. As the criterion for judging the termination, it is possible to use an appropriate criterion generally used in machine learning. For example, the criterion for judging the termination may be whether a specific number of loops has been executed, whether the loss function is equal to or less than a certain value, or whether overlearning has occurred. In addition, the criterion for judging the termination may be whether the performance has reached a sufficient level by evaluating the PSNR (Peak Signal-to-Noise Ratio) or SSIM (Structual Similarity), which are indexes indicating the performance of noise reduction in the inferred image.

[0083] By the processing of steps S411 to S417, a trained model can be created that can generate images that have been subjected to noise reduction processing that suitably reduces noise, even when the characteristics of the radiation images used for the training data differ from those of the radiation images used in actual radiography work.

[0084] In addition, if the entire image cannot be processed at once due to performance such as the memory capacity of the image processing unit 22, the image may be divided into small regions of appropriate size (e.g., 256 × 256 pixels) for processing in the inference processing and learning processing, and then combined after processing.

[0085] As described above, the radiation imaging system 1 according to this embodiment includes the radiation generator 30, the detection unit 10, and the control unit 20. The radiation generator 30 functions as an example of a radiation generating device that generates radiation. The detection unit 10 functions as an example of a first imaging device that detects radiation and generates a radiation image. The control unit 20 functions as an example of an image processing device that applies image processing to the radiation image.

[0086] The control unit 20 according to this embodiment includes an acquisition unit 21 and an image processing unit 22. The acquisition unit 21 functions as an example of an acquisition unit that acquires a first radiographic image of the object O obtained using a first imaging device. The image processing unit 22 functions as an example of an image processing unit that performs image processing of the first radiographic image using a trained model obtained using training data based on a second radiographic image obtained using a second imaging device having characteristics different from the characteristics of the first imaging device. Furthermore, the characteristics of the first radiographic image and the characteristics of the second radiographic image are adjusted so that the characteristics of the first radiographic image and the characteristics of the second radiographic image match. Here, the training data includes the second radiographic image whose image characteristics have been adjusted. Furthermore, the image processing unit 22 performs image processing of the first radiographic image by inputting the first radiographic image whose image characteristics have been adjusted into the trained model.

[0087] The control unit 20 according to this embodiment has such a configuration, and can match the characteristics of the first radiographic image to be processed with the characteristics of the second radiographic image included in the learning data. Therefore, the control unit 20 can perform appropriate image processing even when using a trained model that has trained with learning data obtained by an imaging device with characteristics different from that of the imaging device related to the processing target.

[0088] The characteristics of the first radiographic image and the characteristics of the second radiographic image may include at least one of the pixel size and spatial frequency characteristics of the image. In this case, the control unit 20 may include a conversion processing unit 267. The conversion processing unit 267 may function as an example of a conversion unit that converts the image size of at least one of the first radiographic image and the second radiographic image so that the pixel size of the first radiographic image and the pixel size of the second radiographic image match. In this embodiment, the conversion processing unit 267 may convert the image size of the first radiographic image so that the pixel size of the first radiographic image and the pixel size of the second radiographic image match. However, the conversion processing unit 267 may convert the image sizes of the first radiographic image and the second radiographic image so that the pixel size of the first radiographic image and the pixel size of the second radiographic image match.

[0089] For example, when the pixel size of the first imaging device is smaller than the pixel size of the second imaging device, the conversion processing unit 267 can reduce the image size of the first radiographic image. More specifically, the conversion processing unit 267 can reduce the image size of the first radiographic image so that the pixel size of the first radiographic image matches the pixel size of the second radiographic image. With this configuration, the control unit 20 can match the pixel size of the first radiographic image to be processed with the pixel size of the second radiographic image included in the learning data, and can perform appropriate image processing.

[0090] Moreover, the control unit 20 may further include a frequency adjustment unit 266. The frequency adjustment unit 266 functions as an example of an adjustment unit that adjusts the spatial frequency characteristics of at least one of the first and second radiographic images so that the spatial frequency characteristics of the first and second radiographic images match. In this embodiment, the frequency adjustment unit 266 can adjust the spatial frequency characteristics of the second radiographic image so that the spatial frequency characteristics of the first and second radiographic images match. However, the frequency adjustment unit 266 may convert the spatial frequency characteristics of the first and second radiographic images so that the spatial frequency characteristics of the first and second radiographic images match. With this configuration, the control unit 20 can match the spatial frequency characteristics of the first radiographic image to be processed with the spatial frequency characteristics of the second radiographic image included in the learning data, and can perform appropriate image processing.

[0091] The frequency adjustment unit 266 can adjust the spatial frequency characteristics of one of the first and second radiographic images, whose image size has not been converted by the conversion processing unit 267, to match the spatial frequency characteristics of the other of the first and second radiographic images, whose image size has been converted by the conversion processing unit 267. In this embodiment, the frequency adjustment unit 266 can adjust the spatial frequency characteristics of the second radiographic image, whose image size has not been converted by the conversion processing unit 267, to match the spatial frequency characteristics of the first radiographic image, whose image size has been converted by the conversion processing unit 267. With this configuration, the control unit 20 can match the spatial frequency characteristics of the first radiographic image to be processed and the spatial frequency characteristics of the second radiographic image included in the learning data, taking into account the effect of the image size conversion process on the spatial frequency. Therefore, the control unit 20 can perform more appropriate image processing even when using a trained model that has trained learning data obtained by an imaging device with characteristics different from that of the imaging device related to the processing target.

[0092] The characteristics of the first and second image capture devices may include at least one of pixel size, spatial frequency characteristics, Nyquist frequency, and noise characteristics, where the noise characteristics may include characteristics related to quantum noise influenced by the transfer function (spatial frequency characteristics) of the image capture devices.

[0093] Moreover, the image processing according to this embodiment may include noise reduction processing. In this case, the learning data may include an image obtained by adding artificial noise simulating noise in the first radiographic image whose image characteristics have been adjusted to the second radiographic image. Here, the artificial noise may be based on noise according to the spatial frequency characteristics of the first radiographic image whose image characteristics have been adjusted. In this case, the trained model can learn noise reduction processing corresponding to the characteristics of the radiographic image to be processed, so that the control unit 20 can perform more appropriate image processing.

[0094] The control unit 20 may further include a learning processing unit 261. The learning processing unit 261 may function as an example of a learning processing unit that performs learning processing of a trained model using training data. In this case, the control unit 20 may generate a trained model used to perform appropriate image processing.

[0095] Example 2 In the first embodiment, the frequency adjustment process by the frequency adjuster 266 was performed in the learning process in order to match the characteristics of the radiation image input in the inference process with the characteristics of the radiation image used as learning data in the learning process. In contrast, it is also possible to adopt a configuration in which the frequency adjustment process is performed in the inference process. Hereinafter, as the second embodiment of the present disclosure, a configuration example in which the frequency adjustment process by the frequency adjuster 266 is performed in the inference process will be described with reference to Figs. 5A and 5B. Note that the configuration of the radiation imaging system according to this embodiment is similar to the configuration of the radiation imaging system 1 according to the first embodiment, and therefore the same reference numerals will be used and the description will be omitted. Hereinafter, the radiation imaging system according to this embodiment will be described, focusing on the differences from the radiation imaging system 1 according to the first embodiment.

[0096] Fig. 5A is a process flow of the inference process in the image processing unit 22. Fig. 5B is a schematic diagram of an image generated in the inference process flow and the learning process flow in the image processing unit 22.

[0097] In step S501, the acquisition unit 21 acquires a first radiographic image 51 acquired by a first imaging device having a semiconductor detector with a pixel size P1 [μm] and a modulation transfer function of MTF1. The acquired first radiographic image 51 is sent to the image processing unit 22. The acquisition unit 21 may acquire the first radiographic image from the first imaging device, or from the external storage device 70, etc.

[0098] In step S502, the first radiographic image 51 is subjected to image size conversion processing by the conversion processing unit 267, and is output as a fourth radiographic image 54. The image size conversion processing converts the image size of the first radiographic image 51 so that the pixel size P1 [μm] of the first radiographic image 51 corresponds to the pixel size P2 [μm] of the second radiographic image 52 obtained by using the second imaging device used as learning data. The image size conversion processing may be similar to the image size conversion processing in step S402. In relation to this, the spatial frequency characteristic MTF4 of the fourth radiographic image 54 is obtained by superimposing the influence of the interpolation filter related to the image size conversion processing on the spatial frequency characteristic MTF1.

[0099] In step S503, the frequency adjustment unit 266 performs frequency adjustment processing on the fourth radiographic image 54 to generate a sixth radiographic image 56. This is a correction processing for making the spatial frequency characteristics of the sixth radiographic image 56 closer to the spatial frequency characteristics MTF2 of the second radiographic image 42 used as learning data. The frequency adjustment unit 266 can apply, for example, a compensation filter for making the spatial frequency characteristics MTF4 of the fourth radiographic image 54 approximately equal to the spatial frequency characteristics MTF2 of the second radiographic image 52 used as learning data. Note that if there is no need to perform frequency adjustment processing, step S503 may be skipped.

[0100] In step S504, the inference processing unit 262 performs pre-processing for performing appropriate inference processing on the sixth radiological image 56. The pre-processing may be similar to the pre-processing in step S403.

[0101] In step S505, the inference processing unit 262 performs inference processing using the trained model 55 on the result of the preprocessing in step S504. More specifically, the inference processing unit 262 inputs the result of the preprocessing to the trained model 55, thereby acquiring the inference processing result output from the trained model 55. Note that the trained model 55 only needs to have been trained in advance using the second radiographic image 52 as training data, and does not need to be trained every time a series of image processing is performed. Note that the learning processing according to this embodiment may be similar to the learning processing according to Example 1, except that frequency adjustment processing is not performed.

[0102] In step S506, the inference processing unit 262 performs post-processing on the inference processing result of step S505. The post-processing is to perform reverse processing of the conversions such as various normalizations and smoothings performed in the pre-processing in step S504. By performing the post-processing, the inference processing unit 262 can obtain the third radiological image 53 that has been subjected to noise reduction processing.

[0103] In step S507, the diagnostic image processor 27 performs the various diagnostic image processing described above on the third radiographic image 53 to output an image suitable for diagnosis. The diagnostic image processing in this step is processing suited to the characteristics of the third radiographic image 53, including the pixel size P2 [μm] and the spatial frequency characteristic MTF2.

[0104] The processes in steps S501 to S507 described above also allow the characteristics of the radiation image input in the inference process to match the characteristics of the radiation image used as learning data in the learning process, making it possible to generate a radiation image with appropriately reduced noise.

[0105] The learning process according to this embodiment may be similar to the learning process according to the first embodiment, except that the frequency adjustment process in step S412 is omitted. However, the artificial noise added in step S413 is noise assumed to be present in the sixth radiographic image 56. For example, this noise is noise that takes into consideration changes in frequency characteristics due to the interpolation filter used in the image size conversion process performed by the conversion processing unit 267, in addition to white noise corresponding to the system noise of the sixth radiographic image 56 and quantum noise affected by the spatial frequency characteristic MTF2.

[0106] In this embodiment, the characteristics of the first radiographic image are adjusted so that the characteristics of the first radiographic image and the characteristics of the second radiographic image match. More specifically, the conversion processing unit 267 can convert the image size of the first radiographic image so that the pixel size of the first radiographic image and the pixel size of the second radiographic image match. In addition, the frequency adjustment unit 266 can adjust the spatial frequency characteristics of the first radiographic image so that the spatial frequency characteristics of the first radiographic image and the spatial frequency characteristics of the second radiographic image match. Therefore, the learning data does not include the second radiographic image whose image characteristics have been adjusted, but includes the second radiographic image whose image characteristics have not been adjusted. The image processing unit 22 performs image processing of the first radiographic image by inputting the first radiographic image whose image characteristics have been adjusted into the learned model.

[0107] Even with this configuration, the control unit 20 can match the characteristics of the first radiographic image to be processed with the characteristics of the second radiographic image included in the learning data. Therefore, the control unit 20 can appropriately perform image processing even when using a trained model that has trained with learning data obtained by an imaging device with characteristics different from that of the imaging device related to the processing target.

[0108] Example 3 So far, the case where the pixel size P1 [μm] of the first imaging device is smaller than the pixel size P2 [μm] of the second imaging device that acquires the second radiation image used as learning data for the noise reduction process performed by the noise reduction processing unit 26, i.e., P1 < P2, has been described. However, the present disclosure is not limited to this, and a configuration where P1 > P2 is also possible.

[0109] In the third embodiment of the present disclosure, a case where the pixel size P1 [μm] of the first imaging device is larger than the pixel size P2 [μm] of the second imaging device that acquires the second radiation image used as learning data for the learned model that performs noise reduction processing will be described. Note that since the configuration of the radiation imaging system according to this embodiment is the same as the configuration of the radiation imaging system 1 according to Embodiment 1, the description will be omitted using the same reference numerals. Hereinafter, the radiation imaging system according to this embodiment will be described centering on the differences from the radiation imaging system 1 according to Embodiment 1.

[0110] First, the inference process in Embodiment 3 of the image processing unit 22 will be described with reference to FIGS. 6A and 6B. FIG. 6A is a processing flow of the inference process in the image processing unit 22. FIG. 6B is a schematic diagram of the images generated by the flow of the inference process and the learning process of the image processing unit 22.

[0111] In step S601, the acquisition unit 21 acquires the first radiation image 61 acquired by the first imaging device having a semiconductor detector with a pixel size P1 [μm] and a modulation transfer function of MTF1. The acquired first radiation image 61 is sent to the image processing unit 22. Note that the acquisition unit 21 may acquire the first radiation image from the first imaging device or from an external storage device 70 or the like.

[0112] In step S602, the inference processing unit 262 performs preprocessing for performing appropriate inference processing on the first radiation image 61. The preprocessing may be the same as the preprocessing in step S403.

[0113] In step S603, the inference processing unit 262 performs inference processing on the result of the preprocessing in step S602 using the trained model 64. More specifically, the inference processing unit 262 inputs the result of the preprocessing to the trained model 64, thereby acquiring the inference processing result output from the trained model 64. A learning method for the trained model 64 will be described later. Note that the trained model 64 may be a machine learning model that has been trained in advance, and it is not necessary for learning to be performed for each series of image processing.

[0114] In step S604, the inference processing unit 262 performs post-processing on the inference processing result of step S603. The post-processing is to perform reverse processing of the conversions such as various normalizations and smoothings performed in the pre-processing in step S601. By performing the post-processing, the inference processing unit 262 can obtain a third radiological image 63 that has been subjected to noise reduction processing.

[0115] In step S605, the diagnostic image processor 27 performs the various diagnostic image processing described above on the third radiographic image 63 obtained in step S604, and outputs an image suitable for diagnosis. The diagnostic image processing in this step is processing suited to the characteristics of the third radiographic image 63, which has a pixel size P1 [μm] and a spatial frequency characteristic MTF1.

[0116] Next, a detailed process of the learning processing unit 261 in this embodiment will be described.

[0117] In step S611, the learning data selection unit 263 selects learning data. As the learning data, an arbitrary image is selected from the image group of the second radiographic images 62 acquired by the second imaging device. The second radiographic images 62 used as the learning data may be acquired using the second imaging device. The acquisition unit 21 acquires appropriate radiographic images acquired using the second imaging device as the learning data. The acquisition unit 21 may acquire radiographic images stored in the storage unit 25, or may acquire radiographic images from the external storage device 70 or the like.

[0118] In step S612, the conversion processing unit 267 outputs an image obtained by performing image size conversion on the second radiographic image 62. Here, the image size conversion process converts the image size of the second radiographic image 62 so that the pixel size P2 [μm] of the second radiographic image 62 obtained using the second imaging device corresponds to the pixel size P1 [μm] of the first radiographic image 61 obtained using the first imaging device. As a result, if the width and height of the second radiographic image 62 are width w2 [pixels] and height h2 [pixels], the width and height of the seventh radiographic image 67 are width w2×B [pixels] and height h2×B [pixels], respectively, with a magnification ratio B=P2 / P1. Here, since P1>P2, B is equal to or less than 1, and the image is reduced in size in the image size conversion process.

[0119] In step S613, the frequency adjustment unit 266 performs frequency adjustment processing on the image obtained in step S612 to generate a seventh radiographic image 67. This is a correction processing for making the spatial frequency characteristics of the seventh radiographic image 67 used as learning data closer to the spatial frequency characteristics MTF1 of the first radiographic image 61 used during inference described above. For example, the frequency adjustment unit 266 can apply a compensation filter to the second radiographic image 62 whose image size has been converted so that the spatial frequency characteristics of the second radiographic image 62 whose image size has been converted become approximately equivalent to the spatial frequency characteristics MTF1 of the first radiographic image 61. Note that if frequency adjustment is not necessary, step S613 may be skipped.

[0120] In step S614, the learning data generating unit 264 adds artificial noise to the seventh radiographic image 67. The artificial noise to be added has noise characteristics according to the characteristics to be reduced in the noise reduction process performed during inference. Here, the artificial noise is noise that is assumed to be present in the first radiographic image 61. An example of the artificial noise that can be used is a noise obtained by combining white noise corresponding to the system noise in the first radiographic image 61 and quantum noise affected by the spatial frequency characteristic MTF1.

[0121] In step S614, the training data generation unit 264 can use the seventh radiographic image 67 to which artificial noise has been added as input data for the training data. On the other hand, the training data generation unit 264 can use the seventh radiographic image 67 to which no artificial noise has been added as correct answer data (teacher data) for the training data. The training processing unit 261 can train a machine learning model using a set of these data as training data, thereby generating a trained model that estimates an image with reduced noise from an image containing noise.

[0122] In step S615, the inference processing unit 262 performs appropriate pre-processing on the set of learning data obtained in step S614. This pre-processing is the same as the processing in step S601.

[0123] In step S616, the inference processing unit 262 performs inference processing using a machine learning model on an image of the learning data, which is the seventh radiographic image 67 serving as input data, to which artificial noise has been added, and outputs inference data to which noise reduction processing has been applied.

[0124] In step S617, the parameter update unit 265 compares the inference data with the teacher data and calculates a loss function that quantifies the error between them. The parameter update unit 265 also updates the parameters of the machine learning model (e.g., filter coefficients of a convolutional layer in a CNN) based on the loss function. Examples of the loss function include the mean absolute error (L1loss) and the mean square error (L2loss).

[0125] In step S618, the learning processing unit 261 judges whether to end the learning, and if it is judged that the learning is to be ended, the flow is ended. On the other hand, if the learning processing unit 261 judges that the learning is not to be ended, the flow is returned to step S611, and the flow of steps S611 to S617 is repeated using other data. Note that the criterion for the end judgment may be the same as the criterion for the end judgment regarding step S417.

[0126] The above-described process can also handle cases where the pixel size P1 [μm] of the first imaging device is larger than the pixel size P2 [μm] of the second imaging device that acquires the second radiographic image used as training data for the trained model that performs the noise reduction process. Therefore, in this embodiment, a trained model of the noise reduction process that can appropriately reduce noise even in cases where P1>P2 is created, and the control unit 20 can generate a radiographic image with appropriately reduced noise by using the trained model.

[0127] As described above, in this embodiment, the characteristics of the second radiographic image are adjusted so that the characteristics of the first radiographic image and the characteristics of the second radiographic image match. Therefore, the learning data includes the second radiographic image whose image characteristics have been adjusted. In addition, the image processing unit 22 performs image processing of the first radiographic image by inputting the first radiographic image whose image characteristics have not been adjusted into the learned model.

[0128] Even with this configuration, the control unit 20 can match the characteristics of the first radiographic image to be processed with the characteristics of the second radiographic image included in the learning data. Therefore, the control unit 20 can appropriately perform image processing even when using a trained model that has trained with learning data obtained by an imaging device with characteristics different from that of the imaging device related to the processing target.

[0129] The conversion processing unit 267 according to this embodiment can convert the image size of the second radiographic image so that the pixel size of the first radiographic image and the pixel size of the second radiographic image match. For example, when the pixel size of the first imaging device is larger than the pixel size of the second imaging device, the conversion processing unit 267 can reduce the image size of the second radiographic image. More specifically, the conversion processing unit 267 can reduce the image size of the second radiographic image so that the pixel size of the second radiographic image matches the pixel size of the first radiographic image. With this configuration, the control unit 20 can match the pixel size of the first radiographic image to be processed with the pixel size of the second radiographic image included in the learning data, and can perform appropriate image processing as images having the same Nyquist frequency.

[0130] Moreover, the image processing according to this embodiment may include noise reduction processing. In this case, the learning data may include an image obtained by adding artificial noise simulating noise in a first radiographic image whose image characteristics have not been adjusted to a second radiographic image. Here, the artificial noise may be based on noise according to the spatial frequency characteristics of the first radiographic image whose image characteristics have not been adjusted. In this case, the trained model can learn noise reduction processing corresponding to the characteristics of the radiographic image to be processed, so that the control unit 20 can perform more appropriate image processing.

[0131] As described above, the conversion processing unit 267 can convert the image size of the second radiographic image so that the pixel size of the first radiographic image and the pixel size of the second radiographic image match. Furthermore, the frequency adjustment unit 266 can adjust the spatial frequency characteristics of the second radiographic image so that the spatial frequency characteristics of the first radiographic image and the spatial frequency characteristics of the second radiographic image match. In this embodiment, the control unit 20 includes a learning processing unit 261. Therefore, the control unit 20 can learn a machine learning model using the second radiographic image whose image characteristics have been adjusted by the conversion processing unit 267 and the frequency adjustment unit 266, and can generate a learned model used for performing appropriate image processing.

[0132] (Variation 1) In the above-mentioned first to third embodiments, the noise reduction processing by the noise reduction processing unit 26 has been described as an example of image processing performed by the image processing unit 22. However, the present disclosure is not limited to this, and the above-mentioned configuration can be adopted for any image processing using a machine learning model. For example, the image processing unit 22 may perform super-resolution processing to improve the resolution of the image. In this case, the noise reduction processing unit 26 may function as a super-resolution processing unit that performs super-resolution processing using a trained model. Note that the configuration of the radiation imaging system according to this modified example may be the same as the configuration of the radiation imaging system 1 according to the above-mentioned first to third embodiments, except for the training data.

[0133] In this case, the learning data used in the machine learning can be generated by, for example, a set in which an image with a low resolution is used as input data, and an image with a desired resolution generated by applying a known super-resolution process to the input data is used as correct answer data. The learning data may also be generated by a set in which an image obtained by using an image capture device capable of acquiring an image with a desired resolution is used as correct answer data, and an image with a reduced resolution of the image is used as an input image. Furthermore, the learning data may be generated by a set in which an image obtained by setting a low resolution as the shooting condition is used as input data, and an image obtained by setting a high resolution (desired resolution) as the shooting condition is used as correct answer data. The resolution of the learning data may be set in consideration of the fact that the resolution changes due to an image size adjustment process for the learning data.

[0134] The frequency adjustment process and image size adjustment process for such learning data may be performed in the same manner as described in the above-mentioned embodiments 1 and 3. In addition, in the learning process using such learning data, the process of adding artificial noise may be omitted.

[0135] Furthermore, the noise reduction processing unit may use a trained model that performs image processing combining the noise reduction processing described in the first to third embodiments and the super-resolution processing according to this modification. In this case, the training data may include a set of data in which an image before the super-resolution processing to which artificial noise has been added is used as input data, and an image after the super-resolution processing without adding artificial noise is used as correct answer data. The training process may involve adding artificial noise to a low-resolution image for input data. In this case, the training data may include a set of data in which an image of low resolution to which artificial noise has been added is used as input data, and an image of high resolution to which artificial noise has not been added is used as correct answer data.

[0136] In this modification, the super-resolution processing is performed as the image processing performed by the inference processing unit 262. However, the image processing may be the diagnostic image processing described above. In this case, as the learning data, for example, an image subjected to at least one of gradation processing, emphasis processing, and grid stripe reduction processing may be used as the correct answer data, and the image before the processing may be used as the input data. In this case, the frequency adjustment processing and image size adjustment processing for the learning data may be performed in the same manner as the processing described in the above embodiments 1 and 3. In addition, in the learning processing using such learning data, the artificial noise addition processing may be omitted. In addition, when the diagnostic image processing is performed by the inference processing unit 262, the diagnostic image processing unit 27 and the diagnostic image processing in steps S406, S507, and S605 may be omitted.

[0137] Furthermore, the image processing performed by the inference processing unit 262 may be a combination of the noise reduction processing described in the first embodiment, the super-resolution processing, and the diagnostic image processing described above. In this case, the learning data may include, as correct answer data, images that have been subjected to various image processing other than the noise reduction processing, and images that have been subjected to these processing and have artificial noise added thereto, as input data.

[0138] As described above, the image processing performed by the inference processing unit 262 can include at least one of noise reduction processing, super-resolution processing, gradation processing, enhancement processing, and grid stripe reduction processing. Even with the configuration described in this modified example, it is possible to match the characteristics of the radiographic image input in the inference processing with the characteristics of the radiographic image used as learning data in the learning processing, so that it is possible to generate a radiographic image that has been appropriately subjected to image processing.

[0139] (Variation 2) The machine learning model used by the inference processing unit 262 may be a combination of any layer configuration such as a variational auto-encoder (VAE), a fully convolutional network (FCN), a SegNet, or a DenseNet as a CNN configuration. Also, for example, a configuration using a Vision Transformer (VIT) may be used.

[0140] (Variation 3) Furthermore, the learning data of the various trained models is not limited to data obtained using the detection unit itself that performs actual imaging, but may be data obtained using the same type of detection unit or the same kind of detection unit, depending on the desired configuration. Note that the trained image processing models according to the above-mentioned embodiments and modified examples are considered to extract, for example, the magnitude of the luminance value of the radiation image, the order, inclination, position, distribution, continuity, etc. of the bright and dark areas as part of the feature quantities, and use them in the estimation process related to the generation of the image-processed radiation image.

[0141] The trained models according to the above-described embodiments and modifications can be provided in the control unit 20. The trained models may be configured, for example, by software modules executed by a processor such as a CPU, MPU, GPU, or FPGA, or may be configured by a circuit that performs a specific function such as an ASIC. These trained models may also be provided in another server device connected to the control unit 20. In this case, the control unit 20 can use the trained model by connecting to a server or the like equipped with the trained model via any network such as the Internet. Here, the server equipped with the trained model may be, for example, a cloud server, a fog server, or an edge server. In addition, when configuring a network within a facility, a site including a facility, or an area including multiple facilities to be capable of wireless communication, the reliability of the network may be improved by configuring the network to use radio waves in a dedicated wavelength band that is limited to the facility, site, area, or the like. In addition, the network may be configured by wireless communication that allows high-speed communication, large-capacity communication, low-latency communication, and multiple simultaneous connections.

[0142] (Other Examples) The present invention can also be realized by supplying software (programs) that realize one or more functions of the various embodiments and modifications described above to a system or device via a network or a storage medium, and having a computer in the system or device read and execute the programs. The computer has one or more processors or circuits, and may include multiple separate computers or a network of multiple separate processors or circuits to read and execute computer-executable instructions.

[0143] In this case, the processor or circuitry may include a central processing unit (CPU), a microprocessing unit (MPU), a graphics processing unit (GPU), an application specific integrated circuit (ASIC), or a field programmable gateway (FPGA). The processor or circuitry may also include a digital signal processor (DSP), a data flow processor (DFP), or a neural processing unit (NPU).

[0144] The above disclosure includes the following configurations, methods, and programs. (Configuration 1) an acquisition unit that acquires a first radiographic image of the object to be inspected using a first imaging device; an image processing unit that performs image processing of the first radiographic image using a trained model obtained using training data based on a second radiographic image obtained using a second imaging device having characteristics different from the characteristics of the first imaging device; Equipped with a characteristic of at least one of the first radiographic image and the second radiographic image is adjusted so that the characteristic of the first radiographic image and the characteristic of the second radiographic image match each other; the learning data includes the second radiographic image or a second radiographic image in which characteristics of the image have been adjusted; When the learning data does not include a second radiographic image in which characteristics of the image have been adjusted, the image processing unit performs image processing of the first radiographic image by inputting the first radiographic image in which characteristics of the image have been adjusted to the trained model. (Configuration 2) The image processing device according to configuration 1, wherein, when the learning data includes a second radiographic image in which characteristics of the image have been adjusted, the image processing unit performs image processing of the first radiographic image by inputting the first radiographic image or the first radiographic image in which characteristics of the image have been adjusted to the trained model. (Configuration 3) 3. The image processing device according to configuration 1 or 2, wherein the characteristics of the first radiation image and the characteristics of the second radiation image include at least one of a pixel size and a spatial frequency characteristic of the images. (Configuration 4) 4. The image processing device according to configuration 3, further comprising a conversion unit that converts an image size of at least one of the first radiographic image and the second radiographic image such that a pixel size of the first radiographic image and a pixel size of the second radiographic image match each other. (Configuration 5) 5. The image processing device according to configuration 4, wherein when a pixel size of the first imaging device is smaller than a pixel size of the second imaging device, the conversion section reduces an image size of the first radiation image. (Configuration 6) 5. The image processing device according to configuration 4, wherein when a pixel size of the first imaging device is larger than a pixel size of the second imaging device, the conversion section reduces an image size of the second radiation image. (Configuration 7) 7. The image processing device according to any one of configurations 3 to 6, further comprising an adjustment unit that adjusts spatial frequency characteristics of at least one of the first radiological image and the second radiological image such that the spatial frequency characteristics of the first radiological image and the spatial frequency characteristics of the second radiological image match each other. (Configuration 8) 7. The image processing device according to any one of configurations 4 to 6, further comprising an adjustment unit that adjusts the spatial frequency characteristics of one of the first radiographic image and the second radiographic image, whose image size has not been converted by the conversion unit, to match the spatial frequency characteristics of the other of the first radiographic image and the second radiographic image, whose image size has been converted by the conversion unit. (Configuration 9) 9. The image processing device according to any one of configurations 1 to 8, wherein the characteristics of the first image capture device and the characteristics of the second image capture device include at least one of pixel size, spatial frequency characteristics, Nyquist frequency, and noise characteristics. (Configuration 10) 10. The image processing device according to any one of configurations 1 to 9, wherein the image processing includes at least one of noise reduction processing, super-resolution processing, gradation processing, enhancement processing, and grid stripe reduction processing. (Configuration 11) 11. The image processing device according to claim 10, wherein the learning data includes an image obtained by adding, to the second radiological image, artificial noise that simulates noise in the first radiological image or a first radiological image having characteristics adjusted. (Configuration 12) 12. The image processing device according to configuration 11, wherein the artificial noise is based on noise according to spatial frequency characteristics of the first radiographic image or a first radiographic image having characteristics adjusted therefor. (Configuration 13) A learning processing unit that performs a learning process of a trained model using the training data; a conversion unit that converts an image size of the second radiographic image so that a pixel size of the first radiographic image and a pixel size of the second radiographic image match each other; an adjustment unit that adjusts the spatial frequency characteristics of the second radiographic image so that the spatial frequency characteristics of the first radiographic image and the spatial frequency characteristics of the second radiographic image match each other; 2. The image processing device according to claim 1, further comprising: (Configuration 14) an acquisition unit that acquires a first radiographic image of the object to be inspected using a first imaging device; an image processing unit that performs image processing of the first radiographic image using a trained model obtained using training data based on a second radiographic image obtained using a second imaging device having characteristics different from the characteristics of the first imaging device; Equipped with A characteristic of at least one of the first radiographic image and the second radiographic image is adjusted; When the learning data does not include a second radiographic image in which characteristics of the image have been adjusted, the image processing unit performs image processing of the first radiographic image by inputting the first radiographic image in which characteristics of the image have been adjusted to the trained model. (Configuration 15) an acquisition unit that acquires a first radiographic image of the object to be inspected using a first imaging device; an image processing unit that performs image processing of the first radiographic image using a trained model obtained using training data based on a second radiographic image obtained using a second imaging device having characteristics different from the characteristics of the first imaging device; Equipped with The image processing unit performs image processing of the first radiographic image by inputting the first radiographic image, the image characteristics of which have been adjusted to match the characteristics of the second radiographic image, into the trained model. (Configuration 16) A radiation generating device that generates radiation; An image processing device according to any one of configurations 1 to 15, The first imaging device; A radiation imaging system comprising: (Method 1) acquiring a first radiographic image of the object to be examined using a first imaging device; performing image processing of the first radiographic image using a trained model obtained using training data based on a second radiographic image obtained using a second imaging device having characteristics different from those of the first imaging device; Including, a characteristic of at least one of the first radiographic image and the second radiographic image is adjusted so that the characteristic of the first radiographic image and the characteristic of the second radiographic image match each other; the learning data includes the second radiographic image or a second radiographic image in which characteristics of the image have been adjusted; The image processing method includes, when the learning data does not include a second radiographic image in which characteristics of the image have been adjusted, performing image processing of the first radiographic image by inputting a first radiographic image in which characteristics of the image have been adjusted into the trained model. (Method 2) acquiring a first radiographic image of the object to be examined using a first imaging device; performing image processing of the first radiographic image using a trained model obtained using training data based on a second radiographic image obtained using a second imaging device having characteristics different from those of the first imaging device; Including, A characteristic of at least one of the first radiographic image and the second radiographic image is adjusted; The image processing device, wherein performing the image processing includes, when the learning data does not include a second radiographic image in which characteristics of the image have been adjusted, performing image processing of the first radiographic image by inputting a first radiographic image in which characteristics of the image have been adjusted into the trained model. (Method 3) acquiring a first radiographic image of the object to be examined using a first imaging device; performing image processing of the first radiographic image using a trained model obtained using training data based on a second radiographic image obtained using a second imaging device having characteristics different from those of the first imaging device; Including, The image processing method includes performing image processing of the first radiographic image by inputting the first radiographic image, the image characteristics of which have been adjusted to match the characteristics of the second radiographic image, into the trained model. (Program 1) A program that, when executed by a computer, causes the computer to execute each step of the image processing method according to any one of Methods 1 to 3.

[0145] Although the present disclosure has been described above with reference to the examples and modifications, the present disclosure is not limited to the above examples and modifications. Inventions modified within the scope of the present disclosure and inventions equivalent to the present disclosure are also included in the present disclosure. In addition, the above-mentioned examples and modifications can be appropriately combined within the scope of the present disclosure. [Explanation of symbols]

[0146] 10: detection unit (first imaging device), 20: control unit (image processing device), 21: acquisition unit, 22: image processing unit, O: inspection object

Claims

1. an acquisition unit that acquires a first radiographic image of the object to be inspected using a first imaging device; an image processing unit that performs image processing of the first radiographic image using a trained model obtained using training data based on a second radiographic image obtained using a second imaging device having characteristics different from those of the first imaging device; Equipped with a characteristic of at least one of the first radiographic image and the second radiographic image is adjusted so that the characteristic of the first radiographic image and the characteristic of the second radiographic image match each other; the learning data includes the second radiographic image or a second radiographic image in which characteristics of the image have been adjusted; When the learning data does not include a second radiographic image in which the image characteristics have been adjusted, the image processing unit performs image processing of the first radiographic image by inputting a first radiographic image in which the image characteristics have been adjusted to the trained model.

2. 2. The image processing device according to claim 1, wherein, when the learning data includes a second radiographic image in which characteristics of the image have been adjusted, the image processing unit performs image processing of the first radiographic image by inputting the first radiographic image or the first radiographic image in which characteristics of the image have been adjusted to the trained model.

3. The image processing apparatus of claim 1 , wherein the characteristics of the first radiographic image and the characteristics of the second radiographic image include at least one of a pixel size and a spatial frequency characteristic of the images.

4. The image processing device according to claim 3 , further comprising a conversion unit that converts an image size of at least one of the first radiographic image and the second radiographic image such that a pixel size of the first radiographic image and a pixel size of the second radiographic image match each other.

5. The image processing device according to claim 4 , wherein when a pixel size of the first imaging device is smaller than a pixel size of the second imaging device, the conversion unit reduces an image size of the first radiation image.

6. The image processing device according to claim 4 , wherein when a pixel size of the first imaging device is larger than a pixel size of the second imaging device, the conversion unit reduces an image size of the second radiation image.

7. 4. The image processing device according to claim 3, further comprising an adjustment section that adjusts spatial frequency characteristics of at least one of the first radiographic image and the second radiographic image such that the spatial frequency characteristics of the first radiographic image and the spatial frequency characteristics of the second radiographic image match each other.

8. 5. The image processing device according to claim 4, further comprising an adjustment unit that adjusts spatial frequency characteristics of one of the first radiographic image and the second radiographic image, whose image size has not been converted by the conversion unit, to match spatial frequency characteristics of the other of the first radiographic image and the second radiographic image, whose image size has been converted by the conversion unit.

9. The image processing device according to claim 1 , wherein the characteristics of the first image capture device and the characteristics of the second image capture device include at least one of a pixel size, a spatial frequency characteristic, a Nyquist frequency, and a noise characteristic.

10. The image processing device according to claim 1 , wherein the image processing includes at least one of a noise reduction process, a super-resolution process, a gradation process, an enhancement process, and a grid stripe reduction process.

11. The image processing device according to claim 10 , wherein the learning data includes an image obtained by adding, to the second radiographic image, artificial noise that simulates noise in the first radiographic image or a first radiographic image having characteristics adjusted for the first radiographic image.

12. The image processing apparatus according to claim 11 , wherein the artificial noise is based on noise according to spatial frequency characteristics of the first radiographic image or a first radiographic image having characteristics adjusted therefor.

13. A learning processing unit that performs a learning process of a trained model using the training data; a conversion unit that converts an image size of the second radiographic image so that a pixel size of the first radiographic image and a pixel size of the second radiographic image match each other; an adjustment unit that adjusts the spatial frequency characteristics of the second radiographic image so that the spatial frequency characteristics of the first radiographic image and the spatial frequency characteristics of the second radiographic image match each other; The image processing device according to claim 1 , further comprising:

14. an acquisition unit that acquires a first radiographic image of the object to be inspected using a first imaging device; an image processing unit that performs image processing of the first radiographic image using a trained model obtained using training data based on a second radiographic image obtained using a second imaging device having characteristics different from those of the first imaging device; Equipped with A characteristic of at least one of the first radiographic image and the second radiographic image is adjusted; When the learning data does not include a second radiographic image in which the image characteristics have been adjusted, the image processing unit performs image processing of the first radiographic image by inputting a first radiographic image in which the image characteristics have been adjusted to the trained model.

15. an acquisition unit that acquires a first radiographic image of the object to be inspected using a first imaging device; an image processing unit that performs image processing of the first radiographic image using a trained model obtained using training data based on a second radiographic image obtained using a second imaging device having characteristics different from those of the first imaging device; Equipped with The image processing unit performs image processing of the first radiographic image by inputting the first radiographic image, the image characteristics of which have been adjusted to match the characteristics of the second radiographic image, into the trained model.

16. A radiation generating device that generates radiation; An image processing device according to any one of claims 1 to 15, The first imaging device; A radiation imaging system comprising:

17. acquiring a first radiographic image of the object to be inspected using a first imaging device; performing image processing of the first radiographic image using a trained model obtained using training data based on a second radiographic image obtained using a second imaging device having characteristics different from those of the first imaging device; Including, a characteristic of at least one of the first radiographic image and the second radiographic image is adjusted so that the characteristic of the first radiographic image and the characteristic of the second radiographic image match each other; the learning data includes the second radiographic image or a second radiographic image in which characteristics of the image have been adjusted; The image processing method includes, when the learning data does not include a second radiographic image in which characteristics of the image have been adjusted, performing image processing of the first radiographic image by inputting a first radiographic image in which characteristics of the image have been adjusted into the trained model.

18. acquiring a first radiographic image of the object to be inspected using a first imaging device; performing image processing of the first radiographic image using a trained model obtained using training data based on a second radiographic image obtained using a second imaging device having characteristics different from those of the first imaging device; Including, A characteristic of at least one of the first radiographic image and the second radiographic image is adjusted; The image processing device includes, when the learning data does not include a second radiographic image in which characteristics of the image have been adjusted, performing image processing of the first radiographic image by inputting a first radiographic image in which characteristics of the image have been adjusted into the trained model.

19. acquiring a first radiographic image of the object to be inspected using a first imaging device; performing image processing of the first radiographic image using a trained model obtained using training data based on a second radiographic image obtained using a second imaging device having characteristics different from those of the first imaging device; Including, The image processing method includes performing image processing of the first radiographic image by inputting the first radiographic image, the image characteristics of which have been adjusted to match the characteristics of the second radiographic image, into the trained model.

20. A program which, when executed by a computer, causes the computer to execute each step of the image processing method according to any one of claims 17 to 19.

Images