Image generation apparatus, image generation method, and program

By clustering and sharpening bright spots in indirect photon counting imaging, the device addresses overlapping issues, enabling efficient and cost-effective imaging with higher doses and shorter times.

JP2024165464AActive Publication Date: 2024-11-28JINNO INST
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
JP2023081688
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-17
Publication Date
2024-11-28
Estimated Expiration
2043-05-17

AI Technical Summary

Technical Problem

Indirect photon counting imaging faces challenges when high X-ray doses cause luminescent spots to overlap, making it difficult to determine the center of gravity, necessitating low doses and prolonged imaging times, which is costly due to the need for advanced silicon stacking technology.

Method used

An image generation device and method that extract clusters of bright spots, sharpen each cluster image, and synthesize them to generate a sharpened image, using a neural network for advanced image sharpening.

Benefits of technology

Enables effective photon positioning even with overlapping spots, allowing higher doses and significantly reducing imaging time, enabling inexpensive photon count imaging with reduced frame rates.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024165464000001_ABST
    Figure 2024165464000001_ABST
Patent Text Reader

Abstract

To provide an image generation apparatus, an image generation method, and a program, in which incident positions of photons can be effectively specified also with respect to a plurality of luminescent spots that are mutually overlapped, a larger amount of radiation dosage can be dealt with than in prior art if a frame rate is the same, imaging time can be drastically shortened, and a frame rate required for an optical imaging element can be substantially reduced in medical imaging with limited imaging time, and with which photon count imaging can be realized using an inexpensive optical imaging element.SOLUTION: In an image generation apparatus 300, a cluster of light-emitting luminescent spots generated when each flying particle incident on an imaging device 200 causes a scintillator 211 to emit light, is extracted for each frame image, and sharpening of each cluster image is performed by applying processing to luminescent spot cluster images, and the sharpened cluster image is re-synthesized, in order to generate a sharpened scintillation image or a photon count image of the incident particles.SELECTED DRAWING: Figure 3
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to radiation imaging using an imaging device equipped with a scintillator and an image sensor, for example, scintillation imaging of flying particles such as medical transmitted X-rays, and in particular to an image generating apparatus, an image generating method, and a program that employ indirect photon count imaging. [Background technology]

[0002] As a method for imaging flying particles such as penetrating X-rays, which is typified by medical X-ray transmission imaging, indirect photon count imaging using a scintillator and a high-speed optical imaging element has been proposed (see, for example, Patent Documents 1 and 2).

[0003] In this method, the scintillation light emitted by a single X-ray photon is captured, and the center of gravity of the bright spot is calculated to determine the position of the X-ray incident. This is recorded as data for each X-ray photon, and the measurement is repeated for many X-ray photons to generate a photon count image.

[0004] Unlike conventional direct photon count imaging, this type of indirect photon count imaging has the advantage that it does not require special semiconductor materials in the imaging element and can use optical imaging elements made of silicon, which has good crystallinity and is easy to micro-machine. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2003-279411 A [Patent Document 2] JP 2017-020912 A Summary of the Invention [Problem to be solved by the invention]

[0006] However, the indirect photon count imaging has a problem that when the dose (density of X-rays incident on one frame) increases, the luminescent bright spots come into contact with each other, making it difficult to derive the center of gravity. Therefore, it is necessary to keep the radiation dose during imaging low, and imaging takes a long time.

[0007] A common solution to this problem is to increase the frame rate of optical imaging elements, but for example, to complete medical imaging within 10 seconds, more than 100,000 frames per second are required, which requires advanced silicon layering technology, and this poses the problem of making the imaging elements very expensive.

[0008] The present invention aims to provide an image generating device, an image generating method, and a program that can effectively identify the position of incidence of photons even for multiple overlapping bright spots, thereby enabling a larger dose to be accommodated than before at the same frame rate and dramatically shortening the imaging time, while at the same time significantly reducing the frame rate required for the optical imaging element in medical imaging where imaging time is limited, thereby making it possible to realize photon count imaging using an inexpensive optical imaging element. [Means for solving the problem]

[0009] A first aspect of the present invention is an image generating device that generates a sharpened image associated with imaging information of incident flying particles captured by an imaging device, and includes a cluster extraction unit that extracts clusters of luminous bright spots generated by each particle incident on the imaging device for each frame image, a sharpening processing unit that sharpens each cluster image by processing each of the extracted bright spot cluster images, and a cluster image synthesis unit that generates a sharpened integral image or an incident particle count image by synthesizing the multiple sharpened cluster images.

[0010] A second aspect of the present invention is an image generation method for generating a sharpened image associated with imaging information of incident flying particles by an imaging device, comprising the steps of extracting clusters of luminous bright spots generated by each particle incident on the imaging device for each frame image, sharpening each cluster image by processing each of the extracted bright spot cluster images, and generating a sharpened integral image or an incident particle count image by combining the multiple sharpened cluster images.

[0011] A third aspect of the present invention is a program for causing a computer to execute each step of the image generating method. Effect of the Invention

[0012] In the present invention, a bright spot that emits light due to photons such as X-rays is recognized as a bright spot cluster rather than as an individual bright spot, and processing is performed for each cluster. This makes it possible to effectively identify the incident position of photons even for multiple overlapping bright spots, which means that a larger dose can be handled at the same frame rate, and imaging time can be dramatically reduced. On the other hand, in medical imaging where the imaging time is limited, the frame rate required for the optical imaging element can be significantly reduced, and photon counting imaging can be realized using an inexpensive optical imaging element. [Brief description of the drawings]

[0013] [Figure 1] 1 is a block diagram showing an example of a basic configuration of a radiation imaging apparatus according to a first embodiment of the present invention. [Diagram 2] 1 is a block diagram showing an example of a basic configuration of an imaging device according to a first embodiment of the present invention. [Diagram 3] 1 is a block diagram showing an example of a basic configuration of an image generating device according to a first embodiment of the present invention. [Figure 4] 1A and 1B are diagrams for explaining conventional bright spot detection and photon count image generation processing as a comparative example. [Diagram 5] FIG. 2 is a diagram for explaining the extraction process of bright point clusters according to the first embodiment of the present invention. [Figure 6] FIG. 2 is a diagram showing an image sharpening model using a neural network according to the first embodiment of the present invention. [Figure 7] FIG. 2 is a diagram showing an example of a learning sample of a neural network according to the first embodiment of the present invention. [Figure 8] FIG. 2 is a diagram for explaining an example of creating a learning sample for a neural network according to the first embodiment of the present invention. [Figure 9] FIG. 11 is a diagram showing an image sharpening model using a neural network according to a second embodiment of the present invention. [Figure 10] FIG. 11 is a diagram showing an image sharpening model using a neural network according to a third embodiment of the present invention. [Figure 11] FIG. 13 is a diagram showing an image sharpening model using a neural network according to a fourth embodiment of the present invention. [Figure 12] FIG. 13 is a block diagram showing an example of a basic configuration of an imaging device according to a fifth embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0014] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. An embodiment of the present invention relates to radiation imaging, for example, medical radiography, and in particular, the present embodiment provides an image generating device and an image generating method for flying particles using a scintillator and an image sensor.

[0015] In the image generating device and image generating method of this embodiment, clusters of luminous bright spots generated by each particle entering the imaging device causing the scintillator to emit light are extracted for each frame image, and each cluster image is sharpened by processing the bright spot cluster images. The sharpened cluster images are then recombined to generate a sharpened scintillation image or incident particle count image.

[0016] In this embodiment, a bright spot that emits light due to photons such as X-rays is recognized as a bright spot cluster rather than as an individual bright spot, and processing is performed for each cluster. This makes it possible to effectively identify the incident position of photons even for multiple overlapping bright spots. As a result, it is possible to handle a larger dose than before, for example, nearly 100 times the dose before, at the same frame rate, and dramatically shorten the imaging time. On the other hand, in medical imaging where the imaging time is limited, the frame rate required for the optical imaging element can be significantly reduced, for example to nearly 1 / 100. Therefore, photon counting imaging can be realized using inexpensive optical imaging elements.

[0017] In the embodiments of the present invention, X-rays are exemplified as radiation, but radiation in the present invention may include α-rays, β-rays, γ-rays, etc., which are beams formed by particles (including photons) emitted by radioactive decay, as well as beams having the same or greater energy, such as X-rays, particle beams, and cosmic rays.

[0018] (First embodiment) Below, we will explain an example configuration of a radiation imaging apparatus 100 as a radiation imaging system related to this first embodiment, the configuration of an imaging device 200 which is a component of the radiation imaging apparatus 100, and an example configuration of an image generating apparatus 300, and then we will explain the characteristic image generation function of the image generating apparatus 300. First, an example of the basic configuration of the radiation imaging apparatus 100 will be described.

[0019] (Basic configuration example of radiation imaging device 100) FIG. 1 is a block diagram showing an example of the basic arrangement of a radiation imaging apparatus according to the first embodiment of the present invention.

[0020] The radiation imaging device 100 in Figure 1 is composed of an X-ray tube 110 that irradiates a subject OBJ with X-rays, which are radiation, an imaging device 200 that images the subject OBJ irradiated with the X-rays, which are radiation, an image generating device 300 that generates a sharpened scintillation image (integral image) or an incident particle count image in association with imaging information from the imaging device 200, and a processor 120 as a control unit or processing unit.

[0021] The processor 120 transmits and receives signals and data to and from the imaging device 200, controls the imaging device 200 to perform radiation imaging, and receives the signal obtained thereby from the imaging device 200. This signal includes a measurement value of radiation photons, and the processor 120 generates image data based on this measurement value for displaying an image captured by radiation on a display unit (not shown).

[0022] (Basic configuration example of the imaging device 200) FIG. 2 is a block diagram showing an example of a basic configuration of an imaging device according to the first embodiment of the present invention. FIG. 2 shows a simplified cross-sectional view of an example of an imaging configuration for imaging a bright spot emitting a single X-ray photon.

[0023] The imaging device 200 is formed by bonding together a scintillator 211 that converts radiation into light and an optical imaging element (image sensor) 212 in which multiple pixels having photodetectors that detect light are arranged in a two-dimensional array (matrix). The imaging device 200 is used to image flying particles, for example X-rays, and the flying particles generate scintillation light to generate luminous bright spots.

[0024] The actual measurement is carried out by imaging the luminescent spot by the scintillator 211 which emits one X-ray photon. In FIG. 2, 211 indicates a scintillator, 212 indicates an optical imaging element (image sensor), 213 indicates an imaging surface (pixel array), 214 indicates an adhesive layer, 215 indicates a scintillator plate, 216 indicates an FOP (fiber optic plate), 217 indicates one X-ray photon, and 218 indicates scintillation emission of one X-ray photon.

[0025] The imaging device 200 is formed by bonding a scintillator plate 215 to an optical imaging element (image sensor) 212 via an adhesive layer 214 . The optical imaging element 212 has an imaging surface 213 made up of a pixel array. The scintillator plate 215 is formed by laminating FOPs (fiber optical plates) 216 for protecting the scintillator 211 and the optical imaging element 212 from direct incidence of X-rays. When a single X-ray photon 217 is incident on the scintillator 211, it generates scintillation light 218. The scintillation light 218 is imaged by the optical imaging element 212 via optical diffusion within the scintillator plate 215.

[0026] Typically, this image is recorded as an image of a bright spot emitting a single X-ray photon, and a photon count image is generated by repeating the measurement for many X-ray photons. In contrast to this, in the first embodiment, bright spots that emit light due to X-ray photons are recognized as bright spot clusters rather than as individual bright spots, and processing is carried out for each cluster. The image generating method according to the first embodiment, including this cluster processing, is realized by an image generating device 300 shown in FIG.

[0027] (Basic configuration example of image generating device 300) FIG. 3 is a block diagram showing an example of a basic configuration of an image generating apparatus according to the first embodiment of the present invention.

[0028] The image generating device 300 according to the first embodiment has a function of generating a sharpened scintillation image or an incident particle (X-ray) image associated with imaging information of X-ray photons, which are incident flying particles, captured by the imaging device 200, as shown in FIG. The image generating device 300 recognizes bright spots that emit X-ray photons not as individual bright spots but as bright spot clusters, and performs processing on each cluster.

[0029] The image generating device 300 includes a cluster extracting section 310 , a sharpening processing section 320 , and a cluster image synthesizing section 330 .

[0030] The cluster extraction unit 310 extracts clusters of luminous bright points generated by X-ray photons, which are particles incident on the imaging device 200, for each frame image. The cluster extraction unit 310 extracts a plurality of clusters from each of a plurality of frame images.

[0031] The sharpening processor 320 performs processing on each of the extracted bright spot cluster images, thereby sharpening each cluster image. The sharpening processor 320 can also perform the sharpening process on the cluster images using a neural network that has undergone machine learning.

[0032] The cluster image synthesis unit 330 synthesizes a plurality of sharpened cluster images to generate a sharpened integral image or an incident particle count image.

[0033] Here, we will explain the outline of the image generation method of the first embodiment, in which light-emitting bright spots are treated as bright spot clusters rather than as individual bright spots and processing is performed for each cluster, while comparing it with a comparative example in which cluster processing is not performed.

[0034] First, a comparative example will be described. 4(A) to 4(C) are diagrams for explaining conventional bright spot detection and photon count image generation processing as a comparative example. FIG. 4A is a diagram for explaining the imaging process of the luminescent bright points, FIG. 4B is a diagram for explaining the extraction process of the luminescent bright points, and FIG. 4C is a diagram for explaining the update process of the output image.

[0035] (Bright spot detection and photon count image generation) (A) Imaging of a bright spot In the imaging process of the luminescent bright spots LS, when X-rays are incident on the scintillator, the luminescent bright spots LS are generated in the imaging frame (the frame also contains some noise).

[0036] (B) Extraction of luminescent spots In the process of extracting the luminescent bright spots SL, for example, among a group of pixels whose pixel output exceeds a predetermined threshold, a location where a predetermined number of pixels or more form an island is extracted as the luminescent bright spot SL. This kind of work makes it possible to distinguish noise.

[0037] (C) Update the output image In the process of updating the output image, it is assumed that a photon has been incident on the output center of gravity of the extracted bright point, and the count value of the corresponding pixel in the output image is incremented. By repeating this process for many extracted photons, a photon count image is generated.

[0038] Typically, this image is recorded as an image of a bright spot emitting a single X-ray photon, and a photon count image is generated by repeating the measurement for many X-ray photons. However, as the dose increases, there is a disadvantage that the luminances of the emitted light come into contact with each other, making it difficult to derive the center of gravity. Therefore, in the first embodiment, the bright spots that emit light due to X-ray photons are regarded as bright spot clusters rather than as individual bright spots, and processing is carried out for each cluster. The image generating method according to the first embodiment, including this cluster processing, is realized by an image generating device 300 shown in FIG.

[0039] The extraction process of the bright point clusters LSCLS and the sharpening process of the cluster images performed by the image generating device 300 according to the first embodiment will be described below with reference to specific examples.

[0040] 5(A) to 5(C) are diagrams for explaining the extraction process of bright point clusters according to the first embodiment of the present invention. FIG. 5A is a diagram for explaining the imaging process of the luminescent bright point cluster, FIG. 5B is a diagram for explaining the extraction process of the luminescent bright point cluster, and FIG. 5C is a diagram for explaining the sharpening process of the cluster image.

[0041] (Detection and sharpening of bright point clusters) (A) Imaging of a cluster of luminous bright spots In imaging of the light-emitting bright spot cluster LSCLS, light-emitting bright spots are generated in the imaging frame when X-rays are incident on the scintillator 211. When the incidence density of X-rays in the frame is high, the bright spots come into contact with each other and form a cluster CLS.

[0042] (B) Extraction of luminous clusters In the extraction process of the luminous bright spot cluster LSCLS, as in the comparative example, the cluster extraction unit 310 extracts, for example, a location in a group of pixels whose pixel output exceeds a predetermined threshold, where a predetermined number of pixels or more form an island, as the luminous bright spot cluster LPCLS.

[0043] (C) Sharpening of cluster images In the sharpening process of the cluster images, the sharpening processor 320 sharpens the cluster images using tone modulation or a machine-learned neural network. The sharpened cluster images are linearly added to the corresponding locations in the output image. In conventional bright spot extraction, the output of a bright spot is a binary value for one point in a frame, whereas each output of a cluster image is an image with grayscale.

[0044] (Advantages of cluster extraction) In the first embodiment, the frame image itself is not subjected to the sharpening process, but bright point clusters are extracted from the frame image and the sharpening process is performed for each cluster. This method has the following advantages:

[0045] 1. Noise reduction Once cluster extraction is performed, much of the noise is removed.

[0046] 2. Advanced image sharpening using neural networks When cluster extraction is performed, all or many frames of images are divided into multiple small cluster images, and many cluster images have similar shapes. By inputting these to a neural network that has implemented machine learning, it is possible to perform more advanced image sharpening, and it is possible to obtain an output that is close to, or virtually equivalent to, a photon count image.

[0047] For example, the number of outputs of the neural network can be the same as the number of inputs. For example, it is possible to put a cluster image into a container of 24 pixels x 24 pixels (=576), use it as the input of the neural network, and make the output 24 pixels x 24 pixels (=576). In this case, the output image can have a resolution 1x that of the optical imaging element in Figure 2.

[0048] Also, the number of outputs of the neural network does not necessarily have to be the same as the number of inputs. For example, it is possible to put the cluster image into a container of 24 pixels x 24 pixels (= 576), use it as the input of the neural network, and make the output 48 pixels x 48 pixels (= 2304). In this case, the output image can have four times the resolution of the optical imaging element in Figure 2.

[0049] (Image sharpening by neural network in the first embodiment) FIG. 6 is a diagram showing an image sharpening model using a neural network according to the first embodiment of the present invention.

[0050] In this model, the input is a cluster image 24 x 24 (401). Since the input image is the output value of the sensor, while the label LBL is a photon count image, linear transformations such as intensity scaling and normalization are performed as necessary to ensure consistency between the two and ensure proper learning (402). The output of the net 400 (sharpened image 24 x 24) is a pseudo photon count image in which one X-ray photon corresponds to one pixel output. The net 400 is trained so that the output image of the net 400 approaches the photon count image of the label.

[0051] (An example of an error function during learning) Learning is carried out by changing the weights of the network so as to minimize the value E of the following equation (1). (N is the number of learning samples)

[0052]

number

[0053] The neural network 400 has the following configuration.

[0054] (Example of the configuration of Neural Network 400) Feedforward neural network: 3 stages, Number of input units: 24x24=576, Intermediate output 1:256, Intermediate output 2: 128, Number of output units: 24x24=576

[0055] (Neural network learning sample example) FIG. 7 is a diagram showing an example of a learning sample of a neural network according to the first embodiment of the present invention. The learning sample example in FIG. 7 is formed by 8 pixel×8 pixel input samples SML and labels LBL.

[0056] The output of the neural network for each cluster is a decimal (floating point) number rather than an integer indicating the number of photons, so the gradation remains even after learning. However, for example, in medical X-ray imaging, more than 100 X-ray photons are incident on one pixel area, so the remaining gradation effects are completely buried in the poisson noise and leave no trace. Therefore, the final image obtained by accumulating and synthesizing these images is virtually equivalent to a photon count image, and the same clarity can be obtained.

[0057] (Example of creating a neural network learning sample) Next, an example of creating a learning sample for a neural network will be described. FIG. 8 is a diagram for explaining an example of creating a learning sample for a neural network according to the first embodiment of the present invention.

[0058] In this example, first, a pseudo light-emitting bright point LP is generated (ST501). The artificial bright spots are generated by analyzing a large number of bright spots from a rough bright spot image and deriving the following parameters. The artificial bright spots are generated by randomly arranging scintillation photons within each bright spot according to the parameters.

[0059] (Derived parameters) The average number of scintillation photons per single luminous spot, - Variation in the number of scintillation photons in the luminescent spots, Probability density of scintillation photons based on the light scattering profile from the center of the luminous spot.

[0060] Next, a large number of light emitting bright points LS are randomly arranged on the pseudo frame (ST502). In this example, the learning samples are generated as a pair of input sample SML and label LBL, and the following image is generated. That is, in order to create the input sample SML, a frame image including a cluster of light emitting bright spots LS is generated (ST503). In parallel, a photon count image corresponding to the arrangement positions of the light emitting spots LS is generated in order to create the label LBL (ST504).

[0061] After generating a frame image in step ST503, the light emitting bright point cluster LSCLS is extracted (ST505), and an input sample SML is generated (ST506). On the other hand, after generating a photon count image in step ST504, or after extracting the luminous bright spot clusters LSCLS in step ST505, a photon count image corresponding to each cluster region is extracted (ST507), and a label LBL is created (ST508).

[0062] As described above, according to the first embodiment, in the image generating device 300, clusters of luminous bright spots generated by each flying particle incident on the imaging device 200 causing the scintillator 211 to emit light are extracted for each frame image, and each cluster image is sharpened by processing the bright spot cluster images. The sharpened cluster images are then recombined to generate a sharpened scintillation image or photon count image of the incident particles. In this manner, in the first embodiment, a bright spot that emits light due to photons such as X-rays is recognized as a bright spot cluster rather than as an individual bright spot, and processing is performed for each cluster. This makes it possible to effectively identify the incident position of photons even for multiple overlapping bright spots. As a result, it is possible to handle a larger dose than before, for example, nearly 100 times the dose before, at the same frame rate, and dramatically shorten the imaging time. On the other hand, in medical imaging where the imaging time is limited, the frame rate required for the optical imaging element can be significantly reduced, for example to nearly 1 / 100. Therefore, photon counting imaging can be realized using an inexpensive optical imaging element.

[0063] (Image sharpening by neural network in the second embodiment) FIG. 9 is a diagram showing an image sharpening model using a neural network according to the second embodiment of the present invention.

[0064] The image sharpening model using a neural network according to the second embodiment differs from the image sharpening model according to the first embodiment in the following respects.

[0065] In the second embodiment, a filter process 404 including a convolution filter and a filter output 405 are added to the process of the first embodiment. The convolution filters may be trained in a manner similar to the weights of a neural net. In this example, the neural network 400A is formed of two stages of forward propagation type neural networks.

[0066] (An example of an error function during learning) Learning is performed by changing the weights of the network so as to minimize the value E of the following equation 2. The error function during learning is the same as in the first embodiment. (N is the number of learning samples)

[0067]

number

[0068] According to the second embodiment, it is possible to obtain the same effects as those of the first embodiment described above, and further sharpen the image.

[0069] (Image sharpening by neural network in the third embodiment) FIG. 10 is a diagram showing an image sharpening model using a neural network according to the third embodiment of the present invention.

[0070] The image sharpening model using a neural network according to the third embodiment differs from the image sharpening model according to the first embodiment in the following respects.

[0071] In the third embodiment, two neural networks 400B1 and 400B2 share the same role, and each network is trained separately, which facilitates convergence to an optimal solution. Furthermore, a convolution filter or the like may be added in front of each of the neural networks 400B1 and 400B2. Moreover, when the number of X-ray photons of a cluster is determined to be 1, its output intensity and center of gravity may be recorded and applied to material decomposition (processes 406 and 408).

[0072] In the third embodiment, if the number of X-ray photons in a bright spot cluster is determined to be 1 in process 406, the bright spot cluster can be regarded as a single bright spot generated by the incidence of a single X-ray photon. In this case, the emission intensity of the bright spot reflects the energy of the incident X-ray photon, and by linking this to the bright spot center of gravity, it becomes possible to separately obtain the energy distribution of the transmitted X-rays.

[0073] This is a second piece of information independent of the transmitted X-ray image generated by combining the sharpened cluster images, and while the former reflects the overall dose distribution of the transmitted X-rays, the latter reflects the energy distribution of each transmitted X-ray photon. Such X-ray energy distribution information can be applied to material decomposition of the subject.

[0074] Specifically, data acquisition is performed as follows. First, the bright spot centroid is calculated by calculating the centroid of the extracted bright spot image (e.g., Fig. 5(C)), as in conventional methods. Meanwhile, the emission intensity of the bright spot is derived from the sum of the outputs of pixels within a certain concentric radius from the centroid position, using the original bright spot image (e.g., Fig. 5(A)). Data linking the two is stored separately from the cluster image, and is considered to be true data only when the number of X-ray photons in the bright spot cluster is determined to be 1, and is used for material decomposition, etc.

[0075] An example of an error function during learning of neural network 400B1 is shown below.

[0076]

number

[0077] The output of the neural network 400B2 can also be regarded as a probability density of class selection in multi-class classification (processes 407 and 408). The labels during learning are considered to be the state that appears when class selection is performed N times (N is the number of photons contained in the cluster). For example, the following error function is set from the maximum likelihood estimation.

[0078]

number

[0079] (Image sharpening by neural network in the fourth embodiment) FIG. 11 is a diagram showing an image sharpening model using a neural network according to the fourth embodiment of the present invention.

[0080] The image sharpening model using a neural network according to the fourth embodiment differs from the image sharpening model according to the first embodiment in the following respects.

[0081] In the fourth embodiment, the resolution is increased by a factor of four via neural network 400C. The corresponding learning sample is generated, for example, by increasing the resolution of the frame image on the label side by four times when randomly arranging the pseudo bright points in FIG.

[0082] An example of an error function during training of neural network 400C is shown below.

[0083]

number

[0084] (Fifth embodiment). FIG. 12 is a block diagram showing an example of the basic configuration of an imaging device according to the fifth embodiment of the present invention.

[0085] In FIG. 12, 231 indicates an electrode, 232 indicates a semiconductor layer for photoelectric conversion, 233 indicates a detection circuit, 234 indicates bonding, 235 indicates an imaging surface (array of multiple pixels), 236 indicates an X-ray photon, and 237 indicates photoelectric conversion of the X-ray photon.

[0086] Direct detection of semiconductors is sometimes used for X-ray detection. In many detection devices, the semiconductor layer 232 that performs photoelectric conversion is not separated for each pixel that constitutes the detection circuit. Therefore, electrons photoelectrically converted from one X-ray photon diffuse in the semiconductor layer 232, which is the photoelectric conversion layer, and reach multiple pixels. Here, when the detection circuit detects a signal due to the reaching electrons in multiple gradations, a luminescent bright spot image similar to scintillation is generated. The present invention is also applicable to such cases.

[0087] The present invention can also be realized by executing the following process: That is, software (programs) that realize the functions of the above-described embodiments are supplied to a system or device via a network or various storage media, and the computer (or CPU, MPU, etc.) of the system or device reads and executes the programs. [Explanation of symbols]

[0088] 100···Radiation imaging device, 110···X-ray tube, 120···Processor, 200···Imaging device, 211···Scintillator, 212···Optical imaging element, 213···Imaging surface (pixel array), 214···Adhesive layer, 215···Scintillator plate, 216···FOP (fiber optic plate), 217···One X-ray photon, 218···Scintillation emission of one X-ray photon, 300···Image generating device, 310···Cluster extraction unit, 320···Sharpening processing unit, 330···Cluster image synthesis unit.

Claims

1. An image generating apparatus that generates a sharpened image in association with imaging information of an incident flying particle by an imaging device, A cluster extraction unit that extracts clusters of luminous bright points generated by each particle incident on the imaging device for each frame image; a sharpening processing unit that sharpens each of the extracted bright point cluster images by performing processing on the cluster images; a cluster image synthesis unit for generating a sharpened integral image or an incident particle count image by synthesizing the plurality of sharpened cluster images; 13. An image generating device comprising:

2. The imaging device is A scintillator that converts incident flying particles into light; an optical imaging element in which a plurality of pixels are arranged in a two-dimensional array, the pixels having a photodetector that detects light and is disposed opposite the scintillator; The imaging device is used to image the flying particles, The scintillation light generated by the flying particles creates a luminous spot.

2. The image generating device of claim 1.

3. The cluster extraction unit Extract multiple clusters from multiple frame images 2. The image generating device of claim 1.

4. The cluster extraction unit Among the pixel groups whose pixel output exceeds a predefined threshold, the areas where a prescribed number of pixels form islands are extracted as luminous bright point clusters.

2. The image generating device of claim 1.

5. The sharpening processing unit includes: Extract bright point clusters from the frame image and perform sharpening processing on each cluster.

2. The image generating device of claim 1.

6. The sharpening processing unit includes: The sharpening process of the cluster images is carried out using a neural network that has undergone machine learning.

3. The image generating device of claim 2.

7. 1. An image generating method for generating a sharpened image associated with imaging information of an incident flying particle by an imaging device, comprising: Extracting clusters of luminous points generated by each particle incident on the imaging device for each frame image; sharpening each of the extracted bright point cluster images by processing the cluster images; generating a sharpened integral image or an incident particle count image by combining the sharpened cluster images; An image generating method comprising:

8. A scintillator that converts incident flying particles into light; an optical imaging element arranged in a two-dimensional array and including a plurality of pixels each having a photodetector for detecting light, the imaging device being used to image flying particles; The scintillation light generated by the flying particles creates a luminous spot. The image generating method of claim 7.

9. Multiple clusters are extracted from multiple frame images. The image generating method of claim 7.

10. The sharpening process of the cluster images is carried out using a neural network that has undergone machine learning. The image generating method of claim 8.

11. A program for causing a computer to execute each step of the image generating method according to any one of claims 7 to 10.

Citation Information

Patent Citations

  • Radiographic imaging device, control method therefor, and program

    JP2017020912A

  • Machine learning-based scintillator response modeling for improved spatial resolution in nuclear imaging

    JP2023505377A

  • X-ray photon-counting data correction through deep learning

    US20220308242A1

  • Method and apparatus for energy selective direct electron imaging

    US20230145436A1

  • Radiation imaging device, and radiation imaging method

    WO2022265097A1