Attenuation distribution image generating device, image processing device, radiation tomography system, attenuation distribution image generating method, image processing method

Abstract

A radiation tomography system (1) includes a radiation tomography apparatus (10) and an image processing apparatus (20). The image processing apparatus (20) includes an image reconstruction unit (21), an attenuation distribution image generation unit (22), and an attenuation correction unit (23). The attenuation distribution image generation unit (22) includes a first processing unit (31) and a second processing unit (32). The first processing unit (31) generates and outputs an intermediate image based on an emission scan image using a learned neural network. The second processing unit (32) generates and outputs an attenuation distribution image based on the intermediate image. Thus, a device and method are realized that can generate a more accurate attenuation distribution image based on an emission scan image even if the emission scan image is obtained using a radiation tomography apparatus different from the radiation tomography apparatus used for learning the neural network.

Description

Technical Field

[0001] The present disclosure relates to an apparatus and method for generating an attenuation distribution image, and an apparatus and method for performing attenuation correction of an emission scan image using the attenuation distribution image. Background Art

[0002] Examples of radiation tomography apparatuses capable of acquiring tomographic images of a subject (living body) include PET (Positron Emission Tomography) apparatuses and SPECT (Single Photon Emission Computed Tomography) apparatuses.

[0003] The PET apparatus includes a detector unit comprising multiple small radiation detectors arranged around a measurement space where a subject is placed. The PET apparatus uses coincidence counting to detect photon pairs with an energy of 511 keV, generated by electron-positron pair annihilation within a subject injected with a positron-emitting isotope (RI ray source), and collects this coincidence count information. Based on this collected coincidence count information, a tomographic image representing the spatial distribution of photon pair generation frequencies (i.e., the spatial distribution of the RI ray source) within the measurement space can be reconstructed.

[0004] In this case, the list data, which contains the coincidence count information collected by the PET device and is arranged in time series, is divided into multiple frames in the order in which it was collected. Image reconstruction processing is performed using the data contained in each frame of the list data, thereby obtaining a dynamic PET image composed of tomographic images from multiple frames. PET devices play an important role in the field of nuclear medicine, and can be used to study, for example, biological functions and higher-level brain functions.

[0005] Photons generated within a subject are not only absorbed and weakened when passing through the subject, but are also absorbed and weakened when passing through the examination bed on which the subject is placed. Therefore, in order to obtain more accurate tomographic images, attenuation correction is required.

[0006] For example, a PET device is used to measure (emission scan) a subject administered with a drug containing an RI ray source to obtain a tomographic image (emission scan image). Furthermore, measurements are performed while rotating the RI ray source around the subject, where no drug has been administered, to obtain a tomographic image (transmission scan image). Furthermore, the transmission scan image is used to perform attenuation correction on the emission scan image.

[0007] In addition to using transmission scan images obtained by a PET device through transmission scanning, as described above, attenuation correction of emission scan images can also use X-ray CT scan images obtained by scanning with an X-ray CT device. Therefore, to perform attenuation correction and obtain more accurate tomographic images, the PET device must include a transmission scan mechanism or an X-ray CT scan mechanism. This has become a factor hindering the simplification and cost reduction of PET devices.

[0008] Hereinafter, the attenuation distribution image containing the energy of 511 keV obtained from the transmission scan image and the X-ray CT scan image is referred to as the attenuation distribution image. The attenuation distribution image represents the attenuation distribution of photons within the subject. By using this attenuation distribution image, attenuation correction can be performed on the transmission scan image.

[0009] The techniques described in Non-Patent Documents 1 and 2 use multiple sets of emission scan images and transmission scan images to train a convolutional neural network (CNN). This trained CNN generates an attenuation distribution image based on the emission scan image to be attenuated, and uses this attenuation distribution image to perform attenuation correction on the emission scan image. This technique, using the trained CNN, enables attenuation correction even in PET devices without a transmission scan mechanism or an X-ray CT scan mechanism.

[0010] Prior art literature

[0011] Non-patent literature

[0012] Non-patent literature 1: Fang Liu et al., "A deep learning approach for 18F-FDG PET attenuation correction", EJNMMI Physics 5:24, 2018

[0013] Non-patent document 2: Donghwi Hwang et al., "Improving the Accuracy ofSimultaneously Reconstructed Activity and Attenuation Maps Using DeepLearning", Journal of Nuclear Medicine Vol.59 No.10, pp.1624-1629, 2018 Summary of the Invention

[0014] Problems to be solved by the invention

[0015] The techniques described in Non-Patent Documents 1 and 2 assume that the PET apparatus used to acquire the multiple sets of emission scan images and transmission scan images used for CNN training is the same as the PET apparatus used to acquire the emission scan images to be corrected for attenuation. This technique cannot generate accurate attenuation distribution images for emission scan images acquired using a different PET apparatus than the one used for CNN training, and therefore cannot perform accurate attenuation correction.

[0016] An object of the present invention is to provide an apparatus and method for generating a more accurate attenuation distribution image from an emission scanogram even if the emission scanogram is acquired using a radiation tomography apparatus different from the radiation tomography apparatus used for learning a neural network.

[0017] Means used to solve problems

[0018] A first embodiment of the present invention is an attenuation distribution image generating device. The attenuation distribution image generating device comprises: (1) a first processing unit that inputs an emission scan image of a target for attenuation correction into a neural network that has been trained using emission scan images of a plurality of subjects as input images and subject region extraction images based on the attenuation distribution images as training images, and outputs an intermediate image from the neural network; and (2) a second processing unit that generates an attenuation distribution image by adding an image of an examination bed on which the subject is placed when acquiring the emission scan image of the target for attenuation correction to the intermediate image.

[0019] A second embodiment of the present invention is an attenuation distribution image generating device. The attenuation distribution image generating device comprises: (1) a first processing unit that inputs an emission scan image of a target for attenuation correction into a neural network that has learned the attenuation distribution image using emission scan images of a plurality of subjects as input images and the attenuation distribution image as training images, and outputs an intermediate image from the neural network; and (2) a second processing unit that generates an attenuation distribution image by adding an image of a bed on which the subject is placed when acquiring the emission scan image of the target for attenuation correction to an image of the subject region extracted based on the intermediate image.

[0020] Another embodiment of the present invention is an image processing device comprising: (1) an attenuation distribution image generating device according to the first or second embodiment; and (2) an attenuation correction unit for performing attenuation correction on an emission scan image to be attenuated using the attenuation distribution image generated by the attenuation distribution image generating device.

[0021] Another embodiment of the present invention is a radiation tomography system. The radiation tomography system comprises: (1) a radiation tomography apparatus for acquiring an emission scanogram to be attenuated and corrected; (2) an attenuation distribution image generating apparatus according to the first or second embodiment; and (3) an attenuation correction unit for performing attenuation correction on the emission scanogram to be attenuated and corrected using the attenuation distribution image generated by the attenuation distribution image generating apparatus.

[0022] A first embodiment of the present invention is a method for generating an attenuation distribution image. The method comprises: (1) a first processing step of inputting an emission scan image of a target for attenuation correction into a neural network that has been trained using emission scan images of a plurality of subjects as input images and subject region extraction images based on the attenuation distribution images as training images, and outputting an intermediate image from the neural network; and (2) a second processing step of adding an image of an examination bed on which the subject is placed when acquiring the emission scan image of the target for attenuation correction to the intermediate image to generate the attenuation distribution image.

[0023] A second embodiment of the present invention is a method for generating an attenuation distribution image. The method comprises: (1) a first processing step of inputting an emission scan image of a target for attenuation correction into a neural network that has been trained using emission scan images of a plurality of subjects as input images and the attenuation distribution image as training images, and outputting an intermediate image from the neural network; and (2) a second processing step of adding an image of an examination bed on which the subject is placed when the emission scan image of the target for attenuation correction is acquired to a subject region extraction image based on the intermediate image, thereby generating an attenuation distribution image.

[0024] Another embodiment of the present invention is an image processing method comprising: (1) the attenuation distribution image generation method of the first or second embodiment described above; and (2) an attenuation correction step of performing attenuation correction on an emission scan image to be attenuated using the attenuation distribution image generated by the attenuation distribution image generation method.

[0025] Effects of the Invention

[0026] According to each aspect of the present invention, even if an emission scanogram is acquired using a radiation tomography apparatus different from the radiation tomography apparatus used for learning a neural network, a more accurate attenuation distribution image can be generated from the emission scanogram. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a diagram showing the configuration of the radiation tomography system 1 .

[0028] Figure 21 is a diagram showing the configuration of a learning system 2 for weakening CNN learning of the first processing unit 31 of the distribution image generating unit 22 .

[0029] Figure 3 This is a flowchart of CNN learning in the first method.

[0030] Figure 4 This is a flowchart of attenuation distribution image generation and attenuation correction according to the first embodiment.

[0031] Figure 5 This is a flowchart of the process of generating a subject region extraction image based on the attenuation distribution image (step S12 ).

[0032] Figure 6 It means Figure 5 FIG. 1 is a diagram showing an example of an image generated in each step of the flowchart.

[0033] Figure 7 This is a flowchart of the second method of CNN learning.

[0034] Figure 8 This is a flowchart of attenuation distribution image generation and attenuation correction according to the second embodiment.

[0035] Figure 9 It is the emission scan image of the attenuated correction object.

[0036] Figure 10 This is an X-ray CT scan image (correct image).

[0037] Figure 11 This is the intermediate image generated in step S22 in the first embodiment assuming that the image of the examination bed is not removed in step S12 , and is the intermediate image generated in step S42 in the second embodiment.

[0038] Figure 12 This is the attenuation distribution image generated in step S23 in the first embodiment.

[0039] Figure 13 This is the subject region extraction image generated in step S43 in the second embodiment.

[0040] Figure 14 This is the attenuation distribution image generated in step S43 in the second embodiment. DETAILED DESCRIPTION

[0041] Embodiments of an attenuation distribution image generation device, an image processing device, a radiation tomography system, an attenuation distribution image generation method, and an image processing method will be described in detail below with reference to the accompanying drawings. In the description of the drawings, identical elements are denoted by identical reference numerals, and duplicate descriptions will be omitted. The present invention is not limited to these examples.

[0042] Figure 1 This figure shows the configuration of a radiation tomography system 1. The radiation tomography system 1 includes a radiation tomography apparatus 10 and an image processing device 20. The image processing device 20 includes an image reconstruction unit 21, an attenuation distribution image generator 22, and an attenuation correction unit 23. The attenuation distribution image generator (attenuation distribution image generator) 22 includes a first processing unit 31 and a second processing unit 32.

[0043] The radiation tomography apparatus 10 performs emission scanning and collects list data (emission scan data) used to reconstruct a tomographic image (emission scan image) of a subject. Examples of the radiation tomography apparatus 10 include a PET apparatus and a SPECT apparatus. The following description assumes that the radiation tomography apparatus 10 is a PET apparatus.

[0044] The radiation tomography apparatus 10 includes a detection unit having multiple small radiation detectors arranged around a measurement space where a subject is placed on an examination bed. The radiation tomography apparatus 10 uses a coincidence counting method to detect photon pairs with an energy of 511 keV, generated by electron-positron pair annihilation within a subject injected with a positron-emitting isotope (RI ray source), and accumulates this coincidence count information. The radiation tomography apparatus 10 then outputs this accumulated coincidence count information, arranged in time series, as list data (emission scan data) to the image processing apparatus 20.

[0045] The list data includes identification information and detection time information of a pair of radiation detectors that simultaneously counted photon pairs. The list data may further include energy information of photons detected by each radiation detector and detection time difference information of the pair of radiation detectors.

[0046] A computer including a CPU, RAM, ROM, and a hard disk drive is used as the image processing device 20. The image processing device 20 also includes an input unit (e.g., a keyboard, a mouse) for receiving input from an operator and a display unit (e.g., a liquid crystal display) for displaying images.

[0047] The image processing device 20 reconstructs an emission scan image based on the emission scan data through an image reconstruction unit 21, generates an attenuation distribution image based on the emission scan image through an attenuation distribution image generation unit 22, and performs attenuation correction of the emission scan image based on the attenuation distribution image through an attenuation correction unit 23.

[0048] As a technique for reconstructing an emission scan image based on emission scan data in the image reconstruction unit 21, there are known image reconstruction techniques based on the maximum likelihood expectation maximization (ML-EM) method and a successive approximation block iteration method that is an improvement of the ML-EM method. Also known image reconstruction techniques based on the successive approximation block iteration method include the ordered subset ML-EM (OSEM) method, the row-action maximum likelihood algorithm (RAMLA) method, and the dynamic RAMLA (DRAMA) method.

[0049] The first processing unit 31 of the attenuation distribution image generator 22 receives input of the emission scan image generated by the reconstruction process of the image reconstruction unit 21 and, using a learned neural network, generates and outputs an intermediate image based on the emission scan image. The neural network used here is preferably a convolutional neural network (CNN). The second processing unit 32 receives input of the intermediate image generated by the first processing unit 31 and, based on the intermediate image, generates and outputs an attenuation distribution image. Details of the first and second processing units 31 and 32 will be described later.

[0050] The attenuation correction unit 23 receives the emission scan data and the attenuation distribution image generated by the attenuation distribution image generation unit 22. The attenuation correction unit 23 performs attenuation-corrected image reconstruction based on the attenuation distribution image to generate a corrected tomographic image.

[0051] Figure 2 4 is a diagram showing a configuration of a learning system 2 for using CNN learning in the first processing unit 31 of the attenuated distribution image generating unit 22. The learning system 2 includes a CNN processing unit 41 and an evaluation unit 42.

[0052] When learning a CNN, first, multiple emission scan images and attenuation distribution images of each subject are prepared. The attenuation distribution images prepared here can be transmission scan images obtained by a PET device or X-ray CT scan images obtained by an X-ray CT device.

[0053] The CNN processing unit 41 inputs the emission scan image of each subject as an input image into the CNN, and outputs an output image from the CNN. The evaluation unit 42 uses the attenuation distribution image of the subject (or the subject region extraction image based on the attenuation distribution image (described later)) as a training image and calculates the difference (e.g., L2 norm) between the output image from the CNN processing unit 41 and the training image. The CNN processing unit 41 corrects the parameters of the CNN based on the difference calculated by the evaluation unit 42. By performing this series of processing using the emission scan images and attenuation distribution images of each of the multiple subjects, the CNN can be trained.

[0054] Next, the processing contents of the first processing unit 31 and the second processing unit 32 of the attenuation distribution image generator 22 will be described. There are two methods for these processes. Correspondingly, there are also two methods for the CNN learning process in the learning system 2.

[0055] use Figure 3 and Figure 4 , the first method of CNN learning and attenuated distribution image generation is explained. Figure 3 This is a flowchart of CNN learning in the first method. Figure 4 This is a flowchart of attenuation distribution image generation and attenuation correction according to the first embodiment.

[0056] In the first CNN learning method, in step S11, emission scan images and attenuation distribution images of multiple subjects are prepared. These attenuation distribution images include not only images of the subjects but also images of the examination table on which they are placed, and may also include other unnecessary images. Unnecessary images are those that are not necessary for diagnosing the subject's condition based on tomographic images, such as images of an arm.

[0057] In step S12, the image of the examination bed and unnecessary images are removed from the attenuation distribution images of each of the multiple subjects, and the subject region (e.g., the head region) is extracted from the attenuation distribution images to generate a subject region extraction image. In step S13, the CNN is trained using the emission scan images (input images) and the subject region extraction images (training images) of each of the multiple subjects via the learning system 2.

[0058] In the attenuation distribution image generation of the first embodiment, in step S21, the image reconstruction unit 21 generates an emission scanogram based on emission scanogram data acquired by the radiation tomography apparatus 10. The emission scanogram generated here becomes the target of attenuation correction.

[0059] In the first processing step S22, the emission scanogram to be attenuated is input into the learned CNN by the first processing unit 31, and the CNN outputs an intermediate image. This intermediate image is an attenuation distribution image corresponding to the emission scanogram to be attenuated, and does not include an image of the examination table (the examination table of the radiation tomography apparatus 10) on which the subject was placed when the emission scanogram was acquired.

[0060] In the second processing step S23, the second processing unit 32 adds the image of the couch (the couch of the radiation tomography apparatus 10) on which the subject is placed when the emission scanogram to be attenuated is acquired to the intermediate image, thereby generating an attenuation distribution image including the couch image. In the attenuation correction step S24, the attenuation correction unit 23 uses the attenuation distribution image generated in step S23 to perform attenuation correction on the emission scanogram generated in step S21, thereby generating a tomographic image after attenuation correction.

[0061] In addition, during CNN learning (step S13) and processing by the first processing unit 31 (step S22), it is preferable that the number of pixels of the image input to the CNN matches the size of the subject in the image. Preprocessing for this purpose is preferably performed before step S13 or before step S22.

[0062] In addition, in the process of extracting the subject region from the attenuation distribution image (step S12), if the respective regions of the subject image, the bed image, and the unnecessary image are known in advance in the attenuation distribution image, the subject region extraction image can be easily generated. Figure 5 Flowchart and Figure 6 As shown in the image example, by sequentially performing binarization processing, closing processing, opening processing, maximum value filtering processing, and mask processing on the attenuation distribution image, a subject region extraction image can also be easily generated.

[0063] use Figure 7 and Figure 8 , the second method of CNN learning and attenuated distribution image generation is explained. Figure 7 This is a flowchart of the second method of CNN learning. Figure 8 This is a flowchart of attenuation distribution image generation and attenuation correction according to the second embodiment.

[0064] In the second embodiment of CNN learning, emission scanograms and attenuation distribution images of multiple subjects are prepared in step S31. The attenuation distribution images include not only images of the subjects but also images of the examination bed on which the subjects are placed, and may also include other unnecessary images.

[0065] In step S32 , the learning system 2 causes the CNN to learn using the emission scan images (input images) and attenuation distribution images (training images) of each of the plurality of subjects.

[0066] In the attenuation distribution image generation of the second embodiment, in step S41, the image reconstruction unit 21 generates an emission scanogram based on emission scanogram data acquired by the radiation tomography apparatus 10. The emission scanogram generated here becomes the target of attenuation correction.

[0067] In the first processing step S42, the emission scanogram for attenuation correction is input into the learned CNN via the first processing unit 31, and the CNN outputs an intermediate image. This intermediate image is an attenuation distribution image corresponding to the emission scanogram for attenuation correction, and may include an image of the examination bed on which the subject is placed.

[0068] In the second processing step S43, the second processing unit 32 extracts the subject region from the intermediate image, generating a subject region extraction image. Furthermore, in step S43, the second processing unit 32 adds the image of the couch (the couch of the radiation tomography apparatus 10) on which the subject was placed when the emission scanogram for attenuation correction was acquired to the subject region extraction image, generating an attenuation distribution image including the couch image. In the attenuation correction step S44, the attenuation correction unit 23 uses the attenuation distribution image generated in step S43 to perform attenuation correction on the emission scanogram generated in step S41, generating a tomographic image after attenuation correction.

[0069] In addition, during CNN learning (step S32) and processing by the first processing unit 31 (step S42), it is preferable that the number of pixels of the image input to the CNN and the size of the subject in the image are consistent. Preprocessing for this purpose is preferably performed before step S32 or before step S42. In addition, the process of extracting the subject region from the intermediate image (step S43) in the second method can be performed in the same manner as the process of extracting the subject region from the attenuation distribution image (step S12) in the first method.

[0070] Next, examples will be described. The first example described below corresponds to the first aspect of the above-described embodiment, and the second example corresponds to the second aspect of the above-described embodiment.

[0071] In both the first and second examples, emission scan images and attenuation distribution images from 1091 cases were used for CNN training. These images were acquired using the Hamamatsu Photonics SHR-12000 head PET system (hereinafter referred to as "System A"). System A has a transmission scanning mechanism, and the attenuation distribution images are transmission scan images.

[0072] In addition, in either of the first embodiment and the second embodiment, the emission scan image ( Figure 9 ) are all images obtained using the whole-body PET / CT device SHR-74000 manufactured by Hamamatsu Photonics Co., Ltd. (hereinafter referred to as "device B"). This device B is equipped with an X-ray CT scanning mechanism. The X-ray CT scan images ( Figure 10 ) is used as a correct image only for comparison with the attenuation distribution image generated by the embodiment. The shapes of the examination beds on which the subjects are placed are different between apparatus A and apparatus B.

[0073] Figure 9 It is the emission scan image of the attenuated correction object. Figure 10 This is an X-ray CT scan image (correct image). Figure 11 This is an intermediate image generated in step S22 assuming that the image of the examination bed is not removed in step S12 in the first embodiment. Figure 12 This is the attenuation distribution image generated in step S23 in the first embodiment. Figure 11 This is also the intermediate image generated in step S42 in the second embodiment. Figure 13 This is the subject region extraction image generated in step S43 in the second embodiment. Figure 14 This is the attenuation distribution image generated in step S43 in the second embodiment.

[0074] Even if the shapes of the examination beds of apparatus A and apparatus B are different, the attenuation distribution images ( Figure 12 、 Figure 14 ) also includes the shape of the image of the examination bed and the X-ray CT scan image (correct image, Figure 10 ) are well consistent. Thus, in this embodiment, even if an emission scan image is acquired using a PET device different from the PET device used for neural network learning, a more accurate attenuation distribution image can be generated from the emission scan image. Furthermore, by using this attenuation distribution image, more accurate attenuation correction of the emission scan image can be performed.

[0075] The present invention is not limited to the above-described embodiments and configuration examples and is capable of various modifications. For example, since multiple drugs may be administered to a subject, the CNN learned for each drug administered to the subject during the emission scan may be used in the first process (steps S22 and S42) of learning a CNN for each drug (steps S13 and S32) and obtaining an intermediate image based on the emission scan image of the attenuated correction target.

[0076] Furthermore, in the first process (steps S22 and S42) of learning a CNN using emission scan images and attenuation distribution images obtained by administering a mixture of multiple drugs to a subject (steps S13 and S32), and obtaining an intermediate image based on the emission scan images for attenuation correction, a CNN learned using the mixture of the multiple drugs can also be used. In this case, since the CNN learning (steps S13 and S32) can capture characteristics of the attenuation distribution that are independent of the type of drug, an accurate attenuation distribution image can be generated regardless of the type of drug administered to the subject when acquiring the emission scan images for attenuation correction, or even if the administered drug is unknown.

[0077] The first attenuation distribution image generating device of the above-mentioned embodiment is configured to include: (1) a first processing unit that inputs an emission scan image of an attenuation correction object into a neural network that has been learned using emission scan images of each of a plurality of subjects as input images and subject region extraction images based on the attenuation distribution images as training images, and outputs an intermediate image from the neural network; and (2) a second processing unit that adds an image of an examination bed on which the subject is placed when obtaining the emission scan image of the attenuation correction object to the intermediate image to generate an attenuation distribution image.

[0078] The second attenuation distribution image generating device of the above-mentioned embodiment is configured to include: (1) a first processing unit that inputs an emission scan image of an attenuation correction target into a neural network that has learned the attenuation distribution image using the emission scan images of each of a plurality of subjects as input images and the attenuation distribution image as a training image, and outputs an intermediate image from the neural network; and (2) a second processing unit that generates an attenuation distribution image by adding an image of an examination bed on which the subject is placed when acquiring the emission scan image of the attenuation correction target to an image of a subject region extracted based on the intermediate image.

[0079] The image processing device of the above embodiment comprises: (1) the first or second attenuation distribution image generating device described above; and (2) an attenuation correction unit that uses the attenuation distribution image generated by the attenuation distribution image generating device to perform attenuation correction on the emission scan image of the attenuation correction object.

[0080] The radiation tomography system of the above-mentioned embodiment comprises: (1) a radiation tomography apparatus for obtaining an emission scan image of an object for attenuation correction; (2) the above-mentioned first or second attenuation distribution image generating apparatus; and (3) an attenuation correction unit for performing attenuation correction on the emission scan image of the object for attenuation correction using the attenuation distribution image generated by the attenuation distribution image generating apparatus.

[0081] The first attenuation distribution image generation method of the above-mentioned embodiment is configured to include: (1) a first processing step of inputting an emission scan image of an attenuation correction object into a neural network that has been learned using emission scan images of each of a plurality of subjects as input images and subject region extraction images based on the attenuation distribution images as training images, and outputting an intermediate image from the neural network; and (2) a second processing step of adding an image of an examination bed on which the subject is placed when obtaining the emission scan image of the attenuation correction object to the intermediate image to generate an attenuation distribution image.

[0082] The second attenuation distribution image generation method of the above-mentioned embodiment is configured to include: (1) a first processing step of inputting an emission scan image of an attenuation correction object into a neural network that has learned the attenuation distribution image using the emission scan images of each of a plurality of subjects as input images and the attenuation distribution image as a training image, and outputting an intermediate image from the neural network; and (2) a second processing step of adding an image of an examination bed on which the subject is placed when the emission scan image of the attenuation correction object is obtained to an image of a subject region extracted based on the intermediate image to generate an attenuation distribution image.

[0083] The image processing method of the above embodiment comprises: (1) the above-mentioned first or second attenuation distribution image generation method; and (2) an attenuation correction step, using the attenuation distribution image generated by the attenuation distribution image generation method to perform attenuation correction on the emission scan image of the attenuation correction object.

[0084] Industrial applicability

[0085] The present invention can be used as an apparatus and method for generating a more accurate attenuation distribution image from an emission scanogram even if the emission scanogram is acquired using a radiation tomography apparatus different from the radiation tomography apparatus used for learning a neural network.

[0086] Explanation of symbols

[0087] 1...Radiation tomography system, 2...Learning system, 10...Radiation tomography apparatus, 20...Image processing apparatus, 21...Image reconstruction unit, 22...Attenuation distribution image generation unit, 23...Attenuation correction unit, 31...First processing unit, 32...Second processing unit, 41...CNN processing unit, 42...Evaluation unit.

Claims

1. A device for generating a weakened distribution image, wherein: have: a first processing unit that inputs the emission scanogram to be attenuated and corrected into a neural network that has been trained using emission scanograms of a plurality of subjects as input images and subject region extraction images based on the attenuation distribution image as training images, and outputs an intermediate image from the neural network; and The second processing unit generates an attenuation distribution image by adding an image of an examination bed on which a subject is placed to the intermediate image when acquiring the emission scanogram to be corrected for attenuation.

2. A device for generating a weakened distribution image, wherein: have: a first processing unit that inputs an emission scanogram to be attenuated and corrected into a neural network that has been trained using emission scanograms of a plurality of subjects as input images and attenuation distribution images as training images, and outputs an intermediate image from the neural network; and The second processing unit generates an attenuation distribution image by adding an image of an examination bed on which a subject is placed when acquiring the emission scanogram to be corrected for attenuation to the subject region extraction image based on the intermediate image.

3. An image processing device, wherein: have: The attenuation distribution image generating device according to claim 1 or 2; and The attenuation correction unit performs attenuation correction on the emission scanogram that is the attenuation correction target, using the attenuation distribution image generated by the attenuation distribution image generating device.

4. A radiation tomography system, wherein: have: A radiation tomography apparatus for obtaining an emission scan image of an attenuation correction object; The attenuation distribution image generating device according to claim 1 or 2; and The attenuation correction unit performs attenuation correction on the emission scanogram that is the attenuation correction target, using the attenuation distribution image generated by the attenuation distribution image generating device.

5. A method for generating a weakened distribution image, wherein: have: In a first processing step, an emission scanogram to be attenuated and corrected is inputted into a neural network that has been trained using emission scanograms of a plurality of subjects as input images and subject region extraction images based on the attenuation distribution image as training images, and the neural network outputs an intermediate image. and The second processing step is to generate an attenuation distribution image by adding an image of the examination bed on which the subject is placed when acquiring the emission scanogram to be corrected for attenuation to the intermediate image.

6. A method for generating a weakened distribution image, wherein: have: In a first processing step, an emission scanogram to be attenuated and corrected is input to a neural network that has been trained using emission scanograms of a plurality of subjects as input images and attenuation distribution images as training images, and an intermediate image is output from the neural network. and The second processing step generates an attenuation distribution image by adding an image of the examination bed on which the subject is placed when acquiring the emission scanogram to be corrected for attenuation to the subject region extracted image based on the intermediate image.

7. An image processing method, wherein: have: The attenuation distribution image generation method according to claim 5 or 6; and The attenuation correction step uses the attenuation distribution image generated by the attenuation distribution image generation method to perform attenuation correction on the emission scanogram that is the attenuation correction target.

Images