Cyclic-consistency based dual-domain PET-CT image synthesis method, system and device

Abstract

The present invention provides a method, system and device for cycle-consistent dual-domain PET-CT image synthesis. First, based on the dual-domain cycle-consistent PET / CT synthesis network framework in the image domain and the projection domain, it includes not only the end-to-end network in the image domain, but also the end-to-end network in the projection domain, and the two domains can be connected through the Radon transform and the inverse Radon transform. Secondly, the method for cycle-consistent dual-domain PET-CT image synthesis provided by the present application includes not only the main task of synthesizing CT images from PET images, but also the secondary task of synthesizing PET images from CT images, so as to form several closed-loop synthesis paths in the dual-domain cycle-consistent PET / CT synthesis network framework. Finally, the present application fully utilizes the potential information hidden in the projection domain through the association of the two domains, and more flexibly applies the cycle consistency loss through the closed-loop path, thereby improving the quality of the synthesized CT images.

Description

Technical Field

[0001] The present application relates to the field of image imaging technology, and in particular, to a dual-domain cycle-consistent PET-CT synthesis method, system and device. Background Art

[0002] As the most advanced nuclear medicine imaging technology, positron emission tomography (PET) is widely used clinically. PET can detect abnormal physiological metabolism in a patient's body at the molecular level change stage of the disease, so as to locate and diagnose related lesions. At present, PET has been widely used in the early screening of brain development diseases (such as Alzheimer's disease and epilepsy) and cancer. However, as a functional imaging technology, PET can only provide physiological activity information, and still needs other modalities (such as tomography) to provide structural information for accurate lesion localization. In addition, PET also needs tomography (CT) to provide the ray attenuation parameters of different tissues for its own attenuation correction. Therefore, clinically, PET and CT are often collected together. However, both PET and CT will cause a large amount of radiation hazards to patients. For some patients who need to undergo multiple PET examinations, in order to reduce radiation hazards, we hope to only collect PET images, and then generate corresponding CT images from the PET images through an image generation method for PET attenuation correction and localization.

[0003] Currently, a large number of deep learning models have been applied to the generation of CT images from PET images. However, these models directly adopt an end-to-end manner to learn the mapping relationship between image domains and ignore the physical information hidden in the projection domain, so the performance is not satisfactory. Recent studies have shown that information in another domain (such as the frequency domain, projection domain, etc.) of medical images is also very important for the synthesis task. For example, S. Kevin Zhou et al. proposed using different networks to extract the image domain information and frequency domain information of magnetic resonance for fast image reconstruction and achieved good results. Compared with the method that only uses image domain information, using the dual domain can well mine different levels of information, and then obtain better generated images. More and more work also uses the dual domain to generate images, and the application of the dual domain is becoming more and more extensive in tasks such as removing CT metal artifacts and low-dose PET reconstruction. Therefore, there is an urgent need for a method that can fully combine the dual-domain information of the image domain and the projection domain to improve the quality of synthesized CT images. Summary of the Invention

[0004] In view of the above-mentioned disadvantages of the prior art, the purpose of the present application is to provide a cycle-consistent dual-domain PET-CT image synthesis method, system and device, which are used to solve the technical problem that the quality of the synthesized CT image is limited due to ignoring the potential information hidden in the projection domain in the prior art.

[0005] To achieve the above and other related objectives, the present application provides a cycle-consistent dual-domain PET-CT image synthesis method, and the method includes: obtaining a real PET image in the image domain and inputting it into the first synthesis network in the image domain to output the first synthesized CT image in the image domain, and performing Radon transform on the first synthesized CT image in the image domain to obtain the first cross-domain reconstructed CT image in the projection domain; performing Radon transform on the real PET image in the image domain to obtain the real PET image in the projection domain and inputting it into the third synthesis network in the projection domain to output the second synthesized CT image in the projection domain, and performing inverse Radon transform on the second synthesized CT image in the projection domain to obtain the second cross-domain reconstructed CT image in the image domain; obtaining a real CT image in the image domain and inputting it into the second synthesis network in the image domain to output the first synthesized PET image in the image domain, and performing Radon transform on the first synthesized PET image in the image domain to obtain the first cross-domain reconstructed PET image in the projection domain; performing Radon transform on the real CT image in the image domain to obtain the real CT image in the projection domain and inputting it into the fourth synthesis network in the projection domain to output the second synthesized PET image in the projection domain, and performing inverse Radon transform on the second synthesized PET image in the projection domain to obtain the second cross-domain reconstructed PET image in the image domain; using the real CT image in the image domain and the second cross-domain reconstructed CT image in the image domain as the supervised loss for training the first synthesis network in the image domain; using the real CT image in the projection domain and the first cross-domain reconstructed CT image in the projection domain as the supervised loss for training the third synthesis network in the projection domain; using the real PET image in the image domain and the second cross-domain reconstructed PET image in the image domain as the supervised loss for training the second synthesis network in the image domain; using the real PET image in the projection domain and the first cross-domain reconstructed PET image in the projection domain as the supervised loss for training the fourth synthesis network in the projection domain.

[0006] In an embodiment of the present application, the first synthesis network in the image domain, the second synthesis network in the image domain, the third synthesis network in the projection domain, and the fourth synthesis network in the projection domain form a dual-domain cycle-consistent PET / CT synthesis network framework based on the image domain and the projection domain; the dual-domain cycle-consistent PET / CT synthesis network framework is trained and optimized based on the supervised loss constraint conditions to obtain an optimized first synthesis network in the image domain, thereby improving the quality of the synthesized CT image.

[0007] In an embodiment of the present application, training and optimizing the dual-domain cycle-consistent PET / CT synthesis network framework based on the supervised loss constraint conditions specifically includes: alternately training the first synthesis network in the image domain and the third synthesis network in the projection domain based on the supervised loss constraint conditions, and alternately training the second synthesis network in the image domain and the fourth synthesis network in the projection domain based on the supervised loss constraint conditions to establish a connection between the image domain and the projection domain; alternately training and optimizing the first synthesis network in the image domain and the second synthesis network in the image domain based on the supervised loss constraint conditions to establish a connection between the main task of synthesizing CT images from PET images and the secondary task of synthesizing PET images from CT images.

[0008] In an embodiment of the present application, alternately training the first synthesis network in the image domain and the third synthesis network in the projection domain based on the supervised loss constraint conditions, and alternately training the second synthesis network in the image domain and the fourth synthesis network in the projection domain based on the supervised loss constraint conditions includes: when training the first synthesis network in the image domain, using the real CT image in the image domain as the supervised loss of the first synthesized CT image in the image domain, and using the second cross-domain reconstructed CT image in the image domain as the cross-domain cycle-consistency supervised loss of the first synthesized CT image in the image domain, and its function expression is: When training the third synthesis network in the projection domain, using the real CT image in the projection domain as the supervised loss of the second synthesized CT image in the projection domain, and using the first cross-domain reconstructed CT image in the projection domain as the cross-domain cycle-consistency supervised loss of the second synthesized CT image in the projection domain, and its function expression is: When training the second synthesis network in the image domain, using the real PET image in the image domain as the supervised loss of the first synthesized PET image in the image domain, and using the second cross-domain reconstructed PET image in the image domain as the cross-domain cycle-consistency supervised loss of the first synthesized PET image in the image domain, and its function expression is: When training the fourth synthesis network in the projection domain, using the real PET image in the projection domain as the supervised loss of the second synthesized PET image in the projection domain, and using the first cross-domain reconstructed PET image in the projection domain as the cross-domain consistency supervised loss of the second synthesized PET image in the projection domain, and its function expression is; Wherein, represents the expectation; x CT represents the obtained real CT image in the image domain; x PETThe true PET image representing the acquired image domain; λ1, λ2, λ3, λ4 represent the hyperparameters in the loss function, controlling the importance of the two parts; F represents the forward Radon transform; F -1 represents the inverse Radon transform; represents the first synthesis network in the image domain; represents the third synthesis network in the projection domain; represents the second synthesis network in the image domain; represents the fourth synthesis network in the projection domain; ‖·‖1 is the first-order norm.

[0009] In an embodiment of the present application, the specific process of alternately training and optimizing the first synthesis network in the image domain and the second synthesis network in the image domain includes: 1) Training and optimizing the first synthesis network in the image domain: Inputting the true PET image in the image domain into the first synthesis network in the image domain to output the first synthesized CT image in the image domain; Inputting the first synthesized CT image in the image domain into the second synthesis network in the image domain to obtain a reconstructed PET image; Performing a Radon transform on the first synthesized CT image in the image domain to obtain the first cross-domain reconstructed CT image in the projection domain, and inputting the first cross-domain reconstructed CT image in the projection domain into the fourth synthesis network in the projection domain to obtain the third synthesized PET image in the projection domain, and performing an inverse Radon transform on the third synthesized PET image in the projection domain to obtain the third cross-domain reconstructed PET image in the image domain; 2) Training and optimizing the second synthesis network in the image domain: Inputting the true CT image in the image domain into the second synthesis network in the image domain to output the first synthesized PET image in the image domain; Inputting the first synthesized PET image in the image domain into the first synthesis network in the image domain to obtain a reconstructed CT image; Performing a Radon transform on the first synthesized PET image in the image domain to obtain the first cross-domain reconstructed PET image in the projection domain, and inputting the first cross-domain reconstructed PET image in the projection domain into the third synthesis network in the projection domain to obtain the third synthesized CT image in the projection domain, and performing an inverse Radon transform on the third synthesized CT image in the projection domain to obtain the third cross-domain reconstructed CT image in the image domain.

[0010] In an embodiment of the present application, the supervised loss constraint conditions for alternately training and optimizing the first synthesis network in the image domain and the second synthesis network in the image domain include: when training and optimizing the first synthesis network in the image domain, using the real CT image in the image domain as the supervised loss of the first synthesized CT image in the image domain; using the real PET image in the image domain as the cycle consistency supervised loss of the reconstructed PET image; using the real PET image in the image domain as the cross-domain cycle consistency supervised loss of the third cross-domain reconstructed PET image in the image domain; the specific loss function expression is: When training and optimizing the second synthesis network in the image domain, using the real PET image in the image domain as the supervised loss of the first synthesized PET image in the image domain; using the real CT image in the image domain as the cycle consistency supervised loss of the reconstructed CT image; using the real CT image in the image domain as the cross-domain cycle consistency supervised loss of the third cross-domain reconstructed CT image in the image domain; the specific loss function expression is: Among them, represents the expectation; x CT represents the real CT image obtained in the image domain; x PET represents the real PET image obtained in the image domain; ξ1, ξ2, ξ3, ξ4 represent hyperparameters in the loss function, controlling the importance of the two parts; F represents the forward Radon transform; F -1 represents the inverse Radon transform; represents the first synthesis network in the image domain; represents the third synthesis network in the projection domain; represents the second synthesis network in the image domain; represents the fourth synthesis network in the projection domain; ‖·‖1 is the first-order norm.

[0011] In an embodiment of the present application, the first synthesis network in the image domain, the second synthesis network in the image domain, the third synthesis network in the projection domain, and the fourth synthesis network in the projection domain are respectively any one of a DNN network, a CNN network, a U-Net network, and a GAN network.

[0012] To achieve the above object and other related objects, the present application provides a cycle-consistent dual-domain PET-CT image synthesis system, comprising: an image domain first synthesis module, configured to obtain a real PET image in the image domain and input it into a first synthesis network in the image domain to output a first synthesized CT image in the image domain, and perform a Radon transform on the first synthesized CT image in the image domain to obtain a first cross-domain reconstructed CT image in the projection domain; a projection domain third synthesis module, configured to perform a Radon transform on the real PET image in the image domain to obtain a real PET image in the projection domain and input it into a third synthesis network in the projection domain to output a second synthesized CT image in the projection domain, and perform an inverse Radon transform on the second synthesized CT image in the projection domain to obtain a second cross-domain reconstructed CT image in the image domain; an image domain second synthesis module, configured to obtain a real CT image in the image domain and input it into a second synthesis network in the image domain to output a first synthesized PET image in the image domain, and perform a Radon transform on the first synthesized PET image in the image domain to obtain a first cross-domain reconstructed PET image in the projection domain; a projection domain fourth synthesis module, configured to perform a Radon transform on the real CT image in the image domain to obtain a real CT image in the projection domain and input it into a fourth synthesis network in the projection domain to output a second synthesized PET image in the projection domain, and perform an inverse Radon transform on the second synthesized PET image in the projection domain to obtain a second cross-domain reconstructed PET image in the image domain; a training processing module, configured to use the real CT image in the image domain and the second cross-domain reconstructed CT image in the image domain as a supervised loss for training the first synthesis network in the image domain; use the real CT image in the projection domain and the first cross-domain reconstructed CT image in the projection domain as a supervised loss for training the third synthesis network in the projection domain; use the real PET image in the image domain and the second cross-domain reconstructed PET image in the image domain as a supervised loss for training the second synthesis network in the image domain; use the real PET image in the projection domain and the first cross-domain reconstructed PET image in the projection domain as a supervised loss for training the fourth synthesis network in the projection domain.

[0013] To achieve the above object and other related objects, the present application provides a computer device, comprising: a memory and a processor; the memory is used for storing a computer program; the processor is used for executing the computer program stored in the memory so that the device executes the method as described above.

[0014] To achieve the above object and other related objects, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the cycle-consistent dual-domain PET-CT image synthesis method as described above is implemented.

[0015] In summary, a method, system, and device for dual-domain PET-CT image synthesis based on cycle consistency provided by the present application have the following beneficial effects:

[0016] First, the dual-domain cycle-consistent PET / CT synthesis network framework based on the image domain and the projection domain provided by the present application not only includes an end-to-end network in the image domain but also includes an end-to-end network in the projection domain, and the two domains can be connected through the Radon transform and the inverse Radon transform;

[0017] Second, the method for dual-domain PET-CT image synthesis based on cycle consistency provided by the present application not only includes the main task of synthesizing CT images from PET images but also includes the secondary task of synthesizing PET images from CT images, so that the dual-domain cycle-consistent PET / CT synthesis network framework based on the image domain and the projection domain forms several closed-loop synthesis paths;

[0018] Finally, the present application fully utilizes the potential information hidden in the projection domain through the association of the two domains, and more flexibly applies the cycle consistency loss through the closed-loop path, thereby improving the quality of the synthesized CT images. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It shows a schematic flow chart of a method for dual-domain PET-CT image synthesis based on cycle consistency in an embodiment of the present application.

[0020] Figure 2A It shows a schematic structural diagram of a dual-domain cycle-consistent PET / CT synthesis network framework based on the image domain and the projection domain in an embodiment of the present application.

[0021] Figure 2B It shows a schematic diagram of a strategy for network training based on supervised loss constraint conditions in an embodiment of the present application.

[0022] Figure 3 It shows a schematic module diagram of a dual-domain PET-CT image synthesis system based on cycle consistency in an embodiment of the present application.

[0023] Figure 4 It shows a schematic structural diagram of a computer device in an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] The following describes the embodiments of the present invention through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0025] It should be noted that in the following description, reference is made to the accompanying drawings which describe several embodiments of the present application. It should be understood that other embodiments may also be used and mechanical composition, structure, electrical, and operational changes may be made without departing from the spirit and scope of the present application. The following detailed description should not be considered restrictive, and the scope of the embodiments of the present application is only defined by the claims of the published patent. The terms used herein are only for describing specific embodiments and are not intended to limit the present application. Spatially relative terms, such as "upper", "lower", "left", "right", "below", "beneath", "lower", "above", "upper", etc., may be used in the text to facilitate the description of the relationship between one element or feature shown in the figure and another element or feature.

[0026] Throughout the specification, unless otherwise clearly defined and limited, terms such as "install", "connect", "couple", "fix", "hold" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances.

[0027] Furthermore, as used herein, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly dictates otherwise. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and in the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that the data used in this way can be interchanged where appropriate so that the embodiments described herein can be implemented in an order different from that shown or described herein. It should be further understood that the terms "comprising" and "including" indicate the presence of the stated features, operations, elements, components, items, types, and / or groups, but do not exclude the presence, occurrence, or addition of one or more other features, operations, elements, components, items, types, and / or groups. The terms "or" and "and / or" used herein are interpreted as inclusive, meaning any one or any combination. Thus, "A, B, or C" or "A, B, and / or C" means "any of the following: A; B; C; A and B; A and C; B and C; A, B, and C". An exception to this definition occurs only when the combination of elements, functions, or operations is inherently mutually exclusive in some way.

[0028] In order to make the objectives, technical solutions, and advantages of the present invention more clearly understood, the technical solutions in the embodiments of the present invention will be further described in detail below through the following embodiments in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the invention.

[0029] As Figure 1 shown, a schematic flowchart of a method for synthesizing dual-domain PET-CT images based on cycle consistency in an embodiment of the present application is presented. The method includes the following steps:

[0030] Step S1: Obtain the real PET image in the image domain and input it into the first synthesis network in the image domain to output the first synthesized CT image in the image domain, and perform a Radon transform on the first synthesized CT image in the image domain to obtain the first cross-domain reconstructed CT image in the projection domain.

[0031] It should be noted that the first synthesis network in the image domain is trained with the obtained real PET image in the image domain as the input and the obtained real CT image in the image domain as the output so that it outputs the first synthesized CT image in the image domain.

[0032] Step S2: Perform a Radon transform on the real PET image in the image domain to obtain the real PET image in the projection domain and input it into the third synthesis network in the projection domain to output the second synthesized CT image in the projection domain, and perform an inverse Radon transform on the second synthesized CT image in the projection domain to obtain the second cross-domain reconstructed CT image in the image domain.

[0033] It should be noted that, using the obtained real PET image in the projection domain as the input and the obtained real CT image in the projection domain as the output, the third synthesis network in the projection domain is trained to output the second synthesized CT image in the projection domain.

[0034] Step S3: Obtain the real CT image in the image domain and input it into the second synthesis network in the image domain to output the first synthesized PET image in the image domain, and perform Radon transform on the first synthesized PET image in the image domain to obtain the first cross-domain reconstructed PET image in the projection domain.

[0035] It should be noted that, using the obtained real CT image in the image domain as the input and the obtained real PET image in the image domain as the output, the second synthesis network in the image domain is trained to output the first synthesized PET image in the image domain.

[0036] Step S4: Perform Radon transform on the real CT image in the image domain to obtain the real CT image in the projection domain and input it into the fourth synthesis network in the projection domain to output the second synthesized PET image in the projection domain, and perform inverse Radon transform on the second synthesized PET image in the projection domain to obtain the second cross-domain reconstructed PET image in the image domain.

[0037] It should be noted that, using the obtained real CT image in the projection domain as the input and the obtained real PET image in the projection domain as the output, the fourth synthesis network in the projection domain is trained to output the second synthesized PET image in the projection domain.

[0038] In an embodiment of the present application, the real CT image in the image domain and the second cross-domain reconstructed CT image in the image domain are used as the supervised loss for training the first synthesis network in the image domain; the real CT image in the projection domain and the first cross-domain reconstructed CT image in the projection domain are used as the supervised loss for training the third synthesis network in the projection domain; the real PET image in the image domain and the second cross-domain reconstructed PET image in the image domain are used as the supervised loss for training the second synthesis network in the image domain; the real PET image in the projection domain and the first cross-domain reconstructed PET image in the projection domain are used as the supervised loss for training the fourth synthesis network in the projection domain.

[0039] It should be noted that the Radon transform refers to performing a projection transform on a digital image matrix in the direction of a specified angle ray. The integral operation link of the Radon transform cancels out the brightness fluctuations caused by noise. Through the Radon transform, a CT image or a PET image in the image domain can be transformed into a corresponding CT sine image or PET sine image in the projection domain; through the inverse Radon transform, a CT sine image or a PET sine image in the projection domain can be transformed into a corresponding CT image or PET image in the image domain. That is, a connection can be established between the image domain and the projection domain through the Radon transform and the inverse Radon transform.

[0040] As Figure 2A shown, it is a schematic structural diagram of a dual-domain cycle-consistent PET / CT synthesis network framework in an embodiment of the present application.

[0041] In an embodiment of the present application, the first synthesis network in the image domain, the second synthesis network in the image domain, the third synthesis network in the projection domain, and the fourth synthesis network in the projection domain constitute a dual-domain cycle-consistent PET / CT synthesis network framework; the dual-domain cycle-consistent PET / CT synthesis network framework is trained and optimized based on the supervised loss constraint conditions to obtain an optimized first synthesis network in the image domain, and then a high-quality synthesized CT image is output.

[0042] It should be noted that the first synthesis network in the image domain refers to the network from PET to CT images in the image domain, Figure 2A shown as hereinafter replaced by "N1 network" for description; the second synthesis network in the image domain refers to the network from CT to PET images in the image domain, Figure 2A shown as hereinafter replaced by "N2 network" for description; the third synthesis network in the projection domain refers to the network from PET to CT sine images in the projection domain, Figure 2A shown as hereinafter replaced by "N3 network" for description; the fourth synthesis network in the projection domain refers to the network from CT to PET sine images in the projection domain, Figure 2A shown as hereinafter replaced by "N4 network" for description.

[0043] Among them, the operator F represents the forward projection, that is, the Radon transform, which can transform from the image domain to the projection domain; the operator F -1 represents the filtered back projection, that is, the inverse Radon transform, which can transform from the projection domain to the image domain.

[0044] As Figure 2BAs shown, it is a schematic diagram of the strategy for network training based on supervised loss constraints in an embodiment of the present application.

[0045] In an embodiment of the present application, training and optimizing the dual-domain cycle-consistent PET / CT synthesis network framework based on the supervised loss constraints specifically includes:

[0046] a) Alternately training the first synthesis network in the image domain and the third synthesis network in the projection domain based on the supervised loss constraints, and alternately training the second synthesis network in the image domain and the fourth synthesis network in the projection domain based on the supervised loss constraints, so as to establish a connection between the image domain and the projection domain.

[0047] Combined with Figure 2B in "Stage1: Dual-domain consistency", in stage 1, a connection is established between the image domain and the projection domain based on dual-domain consistency by alternately training network N1 and network N3, and network N2 and network N4 respectively.

[0048] Specifically, when training the first synthesis network in the image domain, the real CT image in the image domain is used as the supervised loss of the first synthesized CT image in the image domain, and the second cross-domain reconstructed CT image in the image domain is used as the cross-domain cycle-consistent supervised loss of the first synthesized CT image in the image domain. Its function expression is:

[0049]

[0050] When training the third synthesis network in the projection domain, the real CT image in the projection domain is used as the supervised loss of the second synthesized CT image in the projection domain, and the first cross-domain reconstructed CT image in the projection domain is used as the cross-domain cycle-consistent supervised loss of the second synthesized CT image in the projection domain. Its function expression is:

[0051]

[0052] When training the second synthesis network in the image domain, the real PET image in the image domain is used as the supervised loss of the first synthesized PET image in the image domain, and the second cross-domain reconstructed PET image in the image domain is used as the cross-domain cycle-consistent supervised loss of the first synthesized PET image in the image domain. Its function expression is:

[0053]

[0054] When training the fourth synthesis network in the projection domain, use the real PET image in the projection domain as the supervision loss of the second synthesized PET image in the projection domain, and use the first cross-domain reconstructed PET image in the projection domain as the cross-domain consistency supervision loss of the second synthesized PET image in the projection domain. Its functional expression is;

[0055]

[0056] Among them, represents the expectation; x CT represents the real CT image in the obtained image domain; x PET represents the real PET image in the obtained image domain; λ1, λ2, λ3, λ4 represent the hyperparameters in the loss function, controlling the importance of the two parts; F represents the forward Radon transform; F -1 represents the inverse Radon transform; represents the first synthesis network in the image domain; represents the third synthesis network in the projection domain; represents the second synthesis network in the image domain; represents the fourth synthesis network in the projection domain; ‖·‖1 is the first-order norm.

[0057] Preferably, the hyperparameters λ1, λ2, λ3, λ4 are set to 0.5, which are the optimal values found through actual tests.

[0058] b) Alternately train and optimize the first synthesis network in the image domain and the second synthesis network in the image domain to establish a connection between the main task of synthesizing CT images from PET images and the sub-task of synthesizing PET images from CT images.

[0059] Combined with Figure 2B in "Stage2: Cycle consistency", Stage 2: By alternately training and optimizing the N1 network and the N2 network, establish a connection between the main task of synthesizing PET→CT images and the sub-task of synthesizing CT→PET images based on the cycle consistency loss and the cross-domain cycle consistency loss.

[0060] In an embodiment of the present application, the specific process of alternately training and optimizing the first synthesis network in the image domain and the second synthesis network in the image domain includes:

[0061] 1) Train and optimize the first synthesis network in the image domain:

[0062] Input the real PET image in the image domain into the first synthesis network in the image domain to output the first synthesized CT image in the image domain;

[0063] Input the first synthetic CT image in the image domain into the second synthetic network in the image domain to obtain a reconstructed PET image;

[0064] Perform a Radon transform on the first synthetic CT image in the image domain to obtain a first cross-domain reconstructed CT image in the projection domain, and input the first cross-domain reconstructed CT image in the projection domain into the fourth synthetic network in the projection domain to obtain a third synthetic PET image in the projection domain. Perform an inverse Radon transform on the third synthetic PET image in the projection domain to obtain a third cross-domain reconstructed PET image in the image domain;

[0065] 2) Train and optimize the second synthetic network in the image domain:

[0066] Input the real CT image in the image domain into the second synthetic network in the image domain to output and obtain the first synthetic PET image in the image domain;

[0067] Input the first synthetic PET image in the image domain into the first synthetic network in the image domain to obtain a reconstructed CT image;

[0068] Perform a Radon transform on the first synthetic PET image in the image domain to obtain a first cross-domain reconstructed PET image in the projection domain, and input the first cross-domain reconstructed PET image in the projection domain into the third synthetic network in the projection domain to obtain a third synthetic CT image in the projection domain. Perform an inverse Radon transform on the third synthetic CT image in the projection domain to obtain a third cross-domain reconstructed CT image in the image domain.

[0069] Specifically, the supervised loss constraint conditions for alternately training and optimizing the first synthetic network and the second synthetic network in the image domain include:

[0070] When training and optimizing the first synthetic network in the image domain, use the real CT image in the image domain as the supervised loss of the first synthetic CT image in the image domain; use the real PET image in the image domain as the cycle consistency supervised loss of the reconstructed PET image; use the real PET image in the image domain as the cross-domain cycle consistency supervised loss of the third cross-domain reconstructed PET image in the image domain. The specific loss function expression is:

[0071]

[0072] When training and optimizing the second synthesis network of the image domain, use the real PET image of the image domain as the supervision loss of the first synthesized PET image of the image domain; use the real CT image of the image domain as the cycle consistency supervision loss of the reconstructed CT image; use the real CT image of the image domain as the cross-domain cycle consistency supervision loss of the third cross-domain reconstructed CT image of the image domain; the specific loss function expression is:

[0073]

[0074] where denotes the expectation; x CT represents the real CT image of the obtained image domain; x PET represents the real PET image of the obtained image domain; ξ1, ξ2, ξ3, ξ4 represent the hyperparameters in the loss function, controlling the importance of the two parts; F represents the forward Radon transform; F -1 represents the inverse Radon transform; represents the first synthesis network of the image domain; represents the third synthesis network of the projection domain; represents the second synthesis network of the image domain; represents the fourth synthesis network of the projection domain; ‖·‖1 is the first-order norm.

[0075] Preferably, the hyperparameters ξ1, ξ2, ξ3, ξ4 are set to 0.5, which are the optimal values found through actual tests.

[0076] In an embodiment of the present application, the first synthesis network of the image domain, the second synthesis network of the image domain, the third synthesis network of the projection domain, and the fourth synthesis network of the projection domain are respectively any one of a DNN network, a CNN network, a U-Net network, and a GAN network. The structure of this network is not limited in this embodiment.

[0077] In addition, quantitative analysis and comparison are carried out with relatively advanced image generation methods in the prior art, including: U-Net, RU-Net, p2pGAN, CycleGAN, MedGAN. By using a PET / CT device to collect paired PET / CT brain images of 65 patients, and selecting 20 two-dimensional PET / CT images for each patient. Among them, 1000 2D images are used as the training set, 60 as the validation set, and the remaining 240 as the test set. Preferably, the PET / CT device uses the United Imaging uExplorer whole-body PET / CT system because of its fast scanning speed, low dose, high detection sensitivity, and low radiation, and can achieve 4D high-definition dynamic imaging of multiple tissues and organs throughout the body. By calculating the structural similarity (SSIM [%]), peak signal-to-noise ratio (PSNR [dB]), and normalized root mean square error (NRMSE [×10 -2 ) to quantitatively compare the effects of images generated by different methods.

[0078] Table 1 Quantitative comparison results of different image generation methods

[0079]

[0080] Among them, the higher the values of the structural similarity (SSIM) and peak signal-to-noise ratio (PSNR), the better the quality of the synthesized image; the lower the value of the normalized root mean square error (NRMSE), the better the quality of the synthesized image. As can be seen from Table 1, the images synthesized by the dual-domain cycle-consistent PET / CT synthesis network framework provided in this application have better quality.

[0081] As Figure 3 shown, it shows a schematic diagram of the modules of a cycle-consistent dual-domain PET-CT image synthesis system in an embodiment of this application. The cycle-consistent dual-domain PET-CT image synthesis system 300 includes:

[0082] The first synthesis module 310 in the image domain is used to obtain the real PET image in the image domain and input it into the first synthesis network in the image domain to output the first synthesized CT image in the image domain, and perform Radon transform on the first synthesized CT image in the image domain to obtain the first cross-domain reconstructed CT image in the projection domain;

[0083] The third synthesis module 320 in the projection domain is used to perform Radon transform on the real PET image in the image domain to obtain the real PET image in the projection domain and input it into the third synthesis network in the projection domain to output the second synthesized CT image in the projection domain, and perform inverse Radon transform on the second synthesized CT image in the projection domain to obtain the second cross-domain reconstructed CT image in the image domain;

[0084] The second synthesis module 330 in the image domain is used to obtain the real CT image in the image domain and input it into the second synthesis network in the image domain to output the first synthesized PET image in the image domain, and perform Radon transform on the first synthesized PET image in the image domain to obtain the first cross-domain reconstructed PET image in the projection domain;

[0085] The fourth synthesis module 340 in the projection domain is used to perform Radon transform on the real CT image in the image domain to obtain the real CT image in the projection domain and input it into the fourth synthesis network in the projection domain to output the second synthesized PET image in the projection domain, and perform inverse Radon transform on the second synthesized PET image in the projection domain to obtain the second cross-domain reconstructed PET image in the image domain;

[0086] The training processing module 350 is used to use the real CT image in the image domain and the second cross-domain reconstructed CT image in the image domain as the supervised loss for training the first synthesis network in the image domain; use the real CT image in the projection domain and the first cross-domain reconstructed CT image in the projection domain as the supervised loss for training the third synthesis network in the projection domain; use the real PET image in the image domain and the second cross-domain reconstructed PET image in the image domain as the supervised loss for training the second synthesis network in the image domain; use the real PET image in the projection domain and the first cross-domain reconstructed PET image in the projection domain as the supervised loss for training the fourth synthesis network in the projection domain.

[0087] It should be understood that the division of each module of the above system is only a division of logical functions. In actual implementation, they can be fully or partially integrated into a physical entity, or physically separated. And these modules can all be implemented in the form of software called by processing elements; they can also all be implemented in the form of hardware; they can also be implemented in such a way that some modules are called by processing elements in the form of software and some modules are implemented in the form of hardware. For example, the training processing module 350 can be a separately established processing element, or can be integrated in a certain chip of the above system. In addition, it can also be stored in the memory of the above system in the form of program code and called and executed by a certain processing element of the above system to perform the functions of the above training processing module 350. The implementation of other modules is similar. In addition, these modules can be fully or partially integrated together or independently implemented. The processing element mentioned here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each of the above modules can be completed by the integrated logic circuit in the processor element or the instruction in the form of software.

[0088] For example, the above-mentioned modules may be one or more integrated circuits configured to implement the above methods, such as: one or more Application Specific Integrated Circuits (ASICs), or, one or more digital signal processors (DSPs), or, one or more Field Programmable Gate Arrays (FPGAs), etc. Again, when a certain module above is implemented in the form of a processing element scheduling program code, the processing element may be a general-purpose processor, such as a Central Processing Unit (CPU) or other processors that can call program code. Again, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0089] As Figure 4 shown, it is a schematic structural diagram of a computer device 400 in an embodiment of the present application. The computer device 400 includes: a memory 410 and a processor 420; the memory 410 is used to store computer instructions; the processor 420 runs the computer instructions to implement as Figure 1 the method described above.

[0090] In some embodiments, the number of the memory 410 and the processor 420 in the computer device 400 may both be one or more, and Figure 4 one is taken as an example for both.

[0091] In an embodiment of the present application, the processor 420 in the computer device 400 will load instructions corresponding to one or more application program processes into the memory 410 according to the steps as Figure 1 described above, and the processor 420 will run the application program stored in the memory 410, so as to implement as Figure 1 the method described above.

[0092] The memory 410 may include a Random Access Memory (RAM), and may also include a non-volatile memory, such as at least one disk memory. The memory 410 stores an operating system and operation instructions, executable modules or data structures, or subsets thereof, or extended sets thereof, where the operation instructions may include various operation instructions for implementing various operations. The operating system may include various system programs for implementing various basic services and processing hardware-based tasks.

[0093] The processor 420 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0094] In some specific applications, the various components of the computer device 400 are coupled together through a bus system, where the bus system may include, in addition to the data bus, a power bus, a control bus, a status signal bus, etc. However, for the sake of clarity, in Figure 4 all kinds of buses are referred to as the bus system.

[0095] In an embodiment of the present application, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the method as Figure 1 described.

[0096] At any possible level of combination of technical details, the present application may be a system, a method, and / or a computer program product. The computer program product may include a computer-readable storage medium, on which computer-readable program instructions are loaded for enabling a processor to implement various aspects of the present application.

[0097] A computer-readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. A computer-readable storage medium may be, for example, (but is not limited to) an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (non-exhaustive list) of the computer-readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disc (DVD), a memory stick, a floppy disk, a mechanically encoded device, such as a punched card or raised structures in grooves storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage medium as used herein is not construed as an instantaneous signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., an optical pulse through an optical fiber cable), or an electrical signal transmitted through a wire.

[0098] The computer-readable program described herein can be downloaded from the computer-readable storage medium to various computing / processing devices, or downloaded to an external computer or external storage device through a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network may include a copper transmission cable, an optical fiber transmission, a wireless transmission, a router, a firewall, a switch, a gateway computer, and / or an edge server. A network adapter or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0099] The computer program instructions for performing the operations of the present application may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, integrated circuit configuration data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, etc., and procedural programming languages such as the "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., by using an Internet service provider to connect through the Internet). In some embodiments, by using the status information of the computer-readable program instructions to customize an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), the electronic circuit can execute the computer-readable program instructions to implement various aspects of the present application.

[0100] In summary, the present application provides a method, system, and device for cycle-consistent dual-domain PET-CT image synthesis. First, the dual-domain cycle-consistent PET / CT synthesis network framework provided by the present application based on the image domain and the projection domain includes not only an end-to-end network in the image domain but also an end-to-end network in the projection domain, and these two domains can be connected through the Radon transform and the inverse Radon transform. Second, the method for cycle-consistent dual-domain PET-CT image synthesis provided by the present application includes not only the main task of synthesizing CT images from PET images but also the secondary task of synthesizing PET images from CT images, so that the dual-domain cycle-consistent PET / CT synthesis network framework based on the image domain and the projection domain forms several closed-loop synthesis paths. Finally, the present application fully utilizes the potential information hidden in the projection domain through the association of the two domains, and more flexibly uses the cycle consistency loss through the closed-loop paths, thereby improving the quality of the synthesized CT images.

[0101] The present application effectively overcomes various disadvantages in the prior art and has high industrial utilization value.

[0102] The above embodiments merely illustrate the principles and effects of the present invention and are not intended to limit the present invention. Any person familiar with this technology can modify or change the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or changes completed by those with ordinary knowledge in the technical field without departing from the spirit and technical ideas disclosed by the present invention should still be covered by the claims of the present invention.

Claims

1. A cyclic-consistency based dual-domain PET-CT image synthesis method, characterized in that, The method includes: Obtain the real PET image in the image domain and input it into the first synthesis network in the image domain to output the first synthesized CT image in the image domain, and perform Radon transform on the first synthesized CT image in the image domain to obtain the first cross-domain reconstructed CT image in the projection domain; Perform Radon transform on the real PET image in the image domain to obtain the real PET image in the projection domain and input it into the third synthesis network in the projection domain to output the second synthesized CT image in the projection domain, and perform inverse Radon transform on the second synthesized CT image in the projection domain to obtain the second cross-domain reconstructed CT image in the image domain; Obtain the real CT image in the image domain and input it into the second synthesis network in the image domain to output the first synthesized PET image in the image domain, and perform Radon transform on the first synthesized PET image in the image domain to obtain the first cross-domain reconstructed PET image in the projection domain; Perform Radon transform on the real CT image in the image domain to obtain the real CT image in the projection domain and input it into the fourth synthesis network in the projection domain to output the second synthesized PET image in the projection domain, and perform inverse Radon transform on the second synthesized PET image in the projection domain to obtain the second cross-domain reconstructed PET image in the image domain; Wherein, the real CT image in the image domain and the second cross-domain reconstructed CT image in the image domain are used as the supervised loss for training the first synthesis network in the image domain; the real CT image in the projection domain and the first cross-domain reconstructed CT image in the projection domain are used as the supervised loss for training the third synthesis network in the projection domain; the real PET image in the image domain and the second cross-domain reconstructed PET image in the image domain are used as the supervised loss for training the second synthesis network in the image domain; the real PET image in the projection domain and the first cross-domain reconstructed PET image in the projection domain are used as the supervised loss for training the fourth synthesis network in the projection domain.

2. The method for synthesizing cyclic-consistent dual-domain PET-CT images according to claim 1, wherein The first synthesis network in the image domain, the second synthesis network in the image domain, the third synthesis network in the projection domain, and the fourth synthesis network in the projection domain constitute a dual-domain cycle-consistent PET / CT synthesis network framework based on the image domain and the projection domain; Train and optimize the dual-domain cycle-consistent PET / CT synthesis network framework based on the supervised loss constraint conditions to obtain the optimized first synthesis network in the image domain, and further obtain a high-quality synthesized CT image.

3. A cyclic-consistency based dual-domain PET-CT image synthesis method according to claim 2, characterized in that, The training and optimization of the dual-domain cycle-consistent PET / CT synthesis network framework based on the supervised loss constraint conditions specifically include: Alternately train the first synthesis network in the image domain and the third synthesis network in the projection domain based on the supervised loss constraint conditions, and alternately train the second synthesis network in the image domain and the fourth synthesis network in the projection domain based on the supervised loss constraint conditions to establish a connection between the image domain and the projection domain; Alternately train and optimize the first synthesis network in the image domain and the second synthesis network in the image domain, so as to establish a connection between the main task of synthesizing CT images from PET images and the sub-task of synthesizing PET images from CT images.

4. The method for synthesizing dual-domain PET-CT images based on cycle consistency according to claim 3, wherein The alternately training the first synthesis network in the image domain and the third synthesis network in the projection domain based on the supervised loss constraint condition, and alternately training the second synthesis network in the image domain and the fourth synthesis network in the projection domain based on the supervised loss constraint condition, includes: When training the first synthesis network in the image domain, use the real CT image in the image domain as the supervised loss of the first synthesized CT image in the image domain, and use the second cross-domain reconstructed CT image in the image domain as the cross-domain cycle consistency supervised loss of the first synthesized CT image in the image domain, and its function expression is: When training the third synthesis network in the projection domain, use the real CT image in the projection domain as the supervised loss of the second synthesized CT image in the projection domain, and use the first cross-domain reconstructed CT image in the projection domain as the cross-domain cycle consistency supervised loss of the second synthesized CT image in the projection domain, and its function expression is: When training the second synthesis network in the image domain, use the real PET image in the image domain as the supervised loss of the first synthesized PET image in the image domain, and use the second cross-domain reconstructed PET image in the image domain as the cross-domain cycle consistency supervised loss of the first synthesized PET image in the image domain, and its function expression is: When training the fourth synthesis network in the projection domain, use the real PET image in the projection domain as the supervised loss of the second synthesized PET image in the projection domain, and use the first cross-domain reconstructed PET image in the projection domain as the cross-domain consistency supervised loss of the second synthesized PET image in the projection domain, and its function expression is; Among them, represents the expectation; x CT represents the true CT image of the acquired image domain; x PET represents the true PET image of the acquired image domain; λ1, λ2, λ3, λ4 represent the hyperparameters in the loss function, controlling the importance of the two parts; F represents the forward Radon transform; F -1 represents the inverse Radon transform; represents the first synthesis network in the image domain; represents the third synthesis network in the projection domain; represents the second synthesis network in the image domain; represents the fourth synthesis network in the projection domain; ‖·‖1 is the first-order norm.

5. The method for synthesizing dual-domain PET-CT images based on cycle consistency according to claim 3, wherein The specific process of alternately training and optimizing the first synthesis network in the image domain and the second synthesis network in the image domain includes: 1) Train and optimize the first synthesis network in the image domain: Input the real PET image in the image domain into the first synthesis network in the image domain to output the first synthesized CT image in the image domain; Input the first synthesized CT image in the image domain into the second synthesis network in the image domain to obtain a reconstructed PET image; Perform a Radon transform on the first synthesized CT image in the image domain to obtain the first cross-domain reconstructed CT image in the projection domain, and input the first cross-domain reconstructed CT image in the projection domain into the fourth synthesis network in the projection domain to obtain the third synthesized PET image in the projection domain, and perform an inverse Radon transform on the third synthesized PET image in the projection domain to obtain the third cross-domain reconstructed PET image in the image domain; 2) Train and optimize the second synthesis network in the image domain: Input the real CT image in the image domain into the second synthesis network in the image domain to output the first synthesized PET image in the image domain; Input the first synthetic PET image in the image domain into the first synthesis network in the image domain to obtain a reconstructed CT image; Perform a Radon transform on the first synthetic PET image in the image domain to obtain a first cross-domain reconstructed PET image in the projection domain, input the first cross-domain reconstructed PET image in the projection domain into the third synthesis network in the projection domain to obtain a third synthetic CT image in the projection domain, and perform an inverse Radon transform on the third synthetic CT image in the projection domain to obtain a third cross-domain reconstructed CT image in the image domain.

6. The method for synthesizing cycle-consistent dual-domain PET-CT images according to claim 5, wherein The supervision loss constraint conditions for alternately training and optimizing the first synthesis network in the image domain and the second synthesis network in the image domain include: When training and optimizing the first synthesis network in the image domain, use the real CT image in the image domain as the supervision loss for the first synthetic CT image in the image domain; use the real PET image in the image domain as the cycle consistency supervision loss for the reconstructed PET image; use the real PET image in the image domain as the cross-domain cycle consistency supervision loss for the first cross-domain reconstructed PET image in the image domain; the specific loss function expression is: When training and optimizing the second synthesis network in the image domain, use the real PET image in the image domain as the supervision loss for the first synthetic PET image in the image domain; use the real CT image in the image domain as the cycle consistency supervision loss for the reconstructed CT image; use the real CT image in the image domain as the cross-domain cycle consistency supervision loss for the third cross-domain reconstructed CT image in the image domain; the specific loss function expression is: Among them, represents the expectation; x CT represents the true CT image of the acquired image domain; x PET represents the true PET image of the acquired image domain; ξ1, ξ2, ξ3, ξ4 represent the hyperparameters in the loss function, controlling the importance of the two parts; F represents the forward Radon transform; F -1 represents the inverse Radon transform; represents the first synthesis network in the image domain; represents the third synthesis network in the projection domain; represents the second synthesis network in the image domain; represents the fourth synthesis network in the projection domain; ‖·‖1 is the first-order norm.

7. The method for synthesizing cyclic-consistent dual-domain PET-CT images according to claim 1, wherein The first synthesis network in the image domain, the second synthesis network in the image domain, the third synthesis network in the projection domain, and the fourth synthesis network in the projection domain are respectively any one of a DNN network, a CNN network, a U-Net network, and a GAN network.

8. A cycle-consistent dual-domain PET-CT image synthesis system, characterized in that, Include: An image domain first synthesis module, configured to obtain a real PET image in the image domain and input it into the first synthesis network in the image domain to output a first synthetic CT image in the image domain, and perform a Radon transform on the first synthetic CT image in the image domain to obtain a first cross-domain reconstructed CT image in the projection domain; A projection domain third synthesis module, configured to perform a Radon transform on the real PET image in the image domain to obtain a real PET image in the projection domain and input it into the third synthesis network in the projection domain to output a second synthetic CT image in the projection domain, and perform an inverse Radon transform on the second synthetic CT image in the projection domain to obtain a second cross-domain reconstructed CT image in the image domain; An image domain second synthesis module, configured to obtain a real CT image in the image domain and input it into the second synthesis network in the image domain to output a first synthetic PET image in the image domain, and perform a Radon transform on the first synthetic PET image in the image domain to obtain a first cross-domain reconstructed PET image in the projection domain; A fourth synthesis module in the projection domain is configured to perform Radon transform on the real CT image in the image domain to obtain the real CT image in the projection domain, and input it into the fourth synthesis network in the projection domain to output the second synthesized PET image in the projection domain, and perform inverse Radon transform on the second synthesized PET image in the projection domain to obtain the second cross-domain reconstructed PET image in the image domain; A training processing module; is configured to use the real CT image in the image domain and the second cross-domain reconstructed CT image in the image domain as the supervised loss for training the first synthesis network in the image domain; use the real CT image in the projection domain and the first cross-domain reconstructed CT image in the projection domain as the supervised loss for training the third synthesis network in the projection domain; use the real PET image in the image domain and the second cross-domain reconstructed PET image in the image domain as the supervised loss for training the second synthesis network in the image domain; use the real PET image in the projection domain and the first cross-domain reconstructed PET image in the projection domain as the supervised loss for training the fourth synthesis network in the projection domain.

9. A computer device, characterized in that, The device includes: a memory and a processor; The memory is configured to store a computer program; the processor is configured to execute the computer program stored in the memory, so that the device executes the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the method for synthesizing cycle-consistent dual-domain PET-CT images according to any one of claims 1 to 7.

Images