Method, apparatus and device for attenuation correction of pet images

Abstract

This invention provides a PET image attenuation correction method, apparatus, and device, comprising: acquiring an unattenuated PET image and inputting it into a first generative adversarial network (GAN) to output a predicted human tissue distribution map; inputting the predicted human tissue distribution map and the unattenuated PET image into a second GAN cascaded with the first GAN to output a corresponding virtual CT image; and using the virtual CT image to perform attenuation correction on the unattenuated PET image to obtain an attenuated PET image. The human bone tissue enhancement generation framework constructed based on the GAN in this application exhibits significantly richer skeletal structures in CT images generated from NAC-PET images; furthermore, the CT images generated based on the human bone tissue enhancement generation framework constructed in this application achieve better results in terms of grayscale similarity and tissue consistency, thereby improving the accuracy of PET attenuation correction.

Description

Technical Field

[0001] This application relates to the field of image imaging technology, and in particular to a method, apparatus and device for PET image attenuation correction. Background Technology

[0002] Positron emission tomography (PET) is an advanced nuclear imaging technique widely used in clinical practice to visualize human tissue metabolism. During PET imaging, a radioactive tracer is injected into the body. While participating in metabolism, it decays and generates gamma photon signals. These emitted photons are received by an external receiver, allowing for the imaging of human metabolism. However, due to photoelectric absorption and Compton scattering, the photon signal attenuates as it travels through human tissue to the external receiver. Therefore, attenuation correction (AC) is necessary to obtain the true distribution of the tracer within the body.

[0003] Conventional PET attenuation correction methods are based on corresponding CT or MR images, indirectly calculating the attenuation map to obtain the corresponding attenuation coefficient for attenuation correction. On PET / CT scanners, the attenuation map can be directly calculated from CT images based on the linear relationship between Hounsfield units (HU) and the 511 keV attenuation coefficient. On PET / MR scanners, since CT images are not available, it is necessary to segment the MR image to identify different tissues and assign a statistical prior coefficient to each tissue to indirectly calculate the attenuation map. However, in some cases, there are no other clinical needs for MR or CT images besides PET attenuation correction. Therefore, if PET can self-correct, the additional cost of scanning MR or CT images can be avoided in some situations.

[0004] Some existing studies have attempted to generate virtual CT images from unattenuated PET (NAC-PET) or, more directly, attenuated PET (AC-PET) based on generative adversarial network (GAN) image generation techniques. With the advancement of technology, the quality of PET images has improved significantly, making it increasingly feasible to directly generate attenuated PET images from unattenuated PET images. Summary of the Invention

[0005] In view of the shortcomings of the prior art described above, the purpose of this application is to provide a PET image attenuation correction method, apparatus and device to solve the technical problem of excessive error in the bone tissue attenuation coefficient of virtual CT images used for PET attenuation correction in the prior art.

[0006] To achieve the above and other related objectives, this application provides a PET image attenuation correction method, the method comprising: acquiring an unattenuation-corrected PET image and inputting it into a first generative adversarial network to output a predicted human tissue distribution map; inputting the predicted human tissue distribution map and the unattenuation-corrected PET image into a second generative adversarial network cascaded with the first generative adversarial network to output a corresponding virtual CT image; and using the virtual CT image to perform attenuation correction on the unattenuation-corrected PET image to obtain an attenuation-corrected PET image.

[0007] In one embodiment of this application, the first generative adversarial network and the second generative adversarial network each include: a generator network constructed using an encoder-decoder network with skip connections, including 4 convolutional layers, 6 residual layers and 2 deconvolutional layers; and a discriminator network including 5 convolutional layers; wherein the convolutional layers and the deconvolutional layers both use 3D convolutional kernels.

[0008] In one embodiment of this application, the step of acquiring an unattenuated PET image and inputting it into a first generative adversarial network (GAN) to output a predicted human tissue distribution map includes: acquiring an unattenuated PET image and inputting it into the generator network of the first GAN to obtain a predicted human tissue distribution map; acquiring a real CT image, performing filtering processing, and performing threshold segmentation based on pixel values ​​to obtain a real human tissue distribution map; inputting the real human tissue distribution map and the predicted human tissue distribution map into the discriminator network of the first GAN, and improving the generation capability of the generator network of the first GAN based on adversarial training, so that the predicted human tissue distribution map output by it is more reasonable.

[0009] In one embodiment of this application, the threshold segmentation of the real CT image based on pixel values ​​includes: defining the region in the CT image with CT values ​​in the range of [-1000HU, -125HU) as the lung region; defining the region in the CT image with CT values ​​in the range of [-125HU, 10HU) as the fat region; defining the region in the CT image with CT values ​​in the range of [10HU, 90HU) as the soft tissue region; and defining the region in the CT image with CT values ​​in the range of [90HU, 1300HU) as the bone region.

[0010] In one embodiment of this application, the step of inputting the predicted human tissue distribution map and the unattenuated PET image into a second generative adversarial network cascaded with the first generative adversarial network to output a corresponding virtual CT image includes: inputting the unattenuated PET image and the predicted human tissue distribution map into the generator network of the second generative adversarial network to obtain a CT prediction image with rich skeletal structure; inputting the real CT image and the CT prediction image with rich skeletal structure into the discriminator network of the second generative adversarial network, and training the generator network of the second generative adversarial network based on the adversarial principle to output a virtual CT image.

[0011] In one embodiment of this application, the first generative adversarial network uses the Dice loss function for segment consistency constraints; the second generative adversarial network uses the mean absolute error loss function for projection consistency constraints.

[0012] In one embodiment of this application, the second generative adversarial network uses the mean absolute error loss function to perform projection consistency constraints, including: performing Radon transform on the virtual CT image and the acquired real CT image respectively to obtain the corresponding virtual CT projection and real CT projection; and using the mean absolute error loss function to perform consistency constraints on the virtual CT projection and the real CT projection.

[0013] To achieve the above and other related objectives, this application provides a PET image attenuation correction device, comprising: a tissue prediction module for acquiring an unattenuated PET image and inputting it into a first generative adversarial network to output a predicted human tissue distribution map; a virtual training module for inputting the predicted human tissue distribution map and the unattenuated PET image into a second generative adversarial network cascaded with the first generative adversarial network to output a corresponding virtual CT image; and an attenuation correction module for using the virtual CT image to perform attenuation correction on the unattenuated PET image to obtain an attenuated PET image.

[0014] To achieve the above and other related objectives, this application provides a computer device, including: a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program stored in the memory to cause the device to perform the method described above.

[0015] To achieve the above and other related objectives, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the PET image attenuation correction method as described above.

[0016] In summary, the PET image attenuation correction method, apparatus, and device provided in this application have the following beneficial effects: the human bone tissue enhancement generative framework (BEGF) constructed based on generative adversarial networks (GAN) in this application has significantly richer bone structures in CT images generated from NAC-PET images; and the CT images generated based on the BEGF network constructed in this application achieve better results in terms of intensity similarity and tissue consistency, thereby improving the accuracy of PET attenuation correction. Attached Figure Description

[0017] Figure 1 The diagram shown is a flowchart illustrating a PET image attenuation correction method according to an embodiment of this application.

[0018] Figure 2 The diagram shows a flowchart of PET image attenuation correction based on a human bone tissue enhancement framework in one embodiment of this application.

[0019] Figure 3 The diagram shows the structure of a first generative adversarial network and a second generative adversarial network in one embodiment of this application.

[0020] Figure 4 This diagram illustrates the qualitative comparison of virtual CT images generated by different methods in one embodiment of this application.

[0021] Figure 5 The diagram shown is a schematic representation of a PET image attenuation correction device according to one embodiment of this application.

[0022] Figure 6 The diagram shown is a structural schematic of a computer device according to an embodiment of this application. Detailed Implementation

[0023] The following specific examples illustrate the implementation of the present invention. 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, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0024] It should be noted that in the following description, reference is made to the accompanying drawings, which illustrate several embodiments of this application. It should be understood that other embodiments may also be used, and changes in mechanical composition, structure, electrical system, and operation may be made without departing from the spirit and scope of this application. The following detailed description should not be considered limiting, and the scope of the embodiments of this application is defined only by the claims of the published patent. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application. Spatially related terms, such as “upper,” “lower,” “left,” “right,” “below,” “below,” “lower part,” “above,” “upper part,” etc., may be used herein to illustrate the relationship between one element or feature shown in the figures and another element or feature.

[0025] Furthermore, as used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context indicates otherwise. The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data used can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. It should be further understood that the terms “comprising,” “including,” indicate the presence of the stated features, operations, elements, components, items, kinds, and / or groups, but do not exclude the presence, occurrence, or addition of one or more other features, operations, elements, components, items, kinds, and / or groups. The terms “or” and “and / or” as used herein are to be interpreted as inclusive, or mean any one or any combination thereof. Thus, “A, B, or C” or “A, B, and / or C” means “any one of: A; B; C; A and B; A and C; B and C; A, B, and C.” An exception to this definition will only occur if the combination of elements, functions, or operations is inherently mutually exclusive in some way.

[0026] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the invention.

[0027] like Figure 1 The diagram shown illustrates a flowchart of a PET image attenuation correction method according to an embodiment of this application. The method includes the following steps:

[0028] Step S1: Acquire unattenuated PET images and input them into the first generative adversarial network to output a more reasonable predicted map of human tissue distribution.

[0029] In one embodiment of this application, step S1 specifically includes:

[0030] 1) Acquire unattenuated PET images and input them into the generator network of the first generative adversarial network to obtain a predicted map of human tissue distribution.

[0031] Specifically, the generator network of the first generative adversarial network is trained to output a predicted human tissue distribution map by taking the obtained non-attenuation corrected PET image as input and the obtained real human tissue distribution map as output.

[0032] 2) Obtain real CT images, perform filtering processing, and perform threshold segmentation based on pixel values ​​to obtain a real human tissue distribution map.

[0033] It should be noted that the acquired real CT images are filtered to suppress impact noise. Gaussian filtering is preferred because the weights of the Gaussian filter matrix exhibit a Gaussian decay characteristic as the distance from the center pixel increases. Therefore, its effect on pixels farther away from the operator is smaller, thus preserving the image's edge features to some extent. By adjusting the Gaussian smoothing parameters, a trade-off can be achieved between excessive blurring and undersmoothing of image features. Specifically, Gaussian filtering is a process of weighted averaging across the entire image; the value of each pixel is obtained by weighted averaging of its own value and the values ​​of other pixels in its neighborhood. Gaussian filtering is a type of linear smoothing filter used to eliminate Gaussian noise.

[0034] In one embodiment of this application, the threshold segmentation of the real CT image based on pixel values ​​includes: defining the region in the CT image with CT values ​​in the range of [-1000HU, -125HU) as the lung region; defining the region in the CT image with CT values ​​in the range of [-125HU, 10HU) as the fat region; defining the region in the CT image with CT values ​​in the range of [10HU, 90HU) as the soft tissue region; and defining the region in the CT image with CT values ​​in the range of [90HU, 1300HU) as the bone region.

[0035] It should be noted that the CT value is the relative density of human tissue, commonly expressed in Henle units (HU), hence the name HU value. It reflects the degree to which human tissue absorbs X-rays. The HU value is independent of the equipment used; different ranges can represent different organs. The absorption rate of water is used as a reference, i.e., HU = 0 for water. Values ​​with an attenuation coefficient greater than that of water are positive, and those less are negative. The HU values ​​of bone cortex and air are used as the upper and lower limits, respectively.

[0036] The pixel values ​​of the acquired CT images are converted into HU values, and the CT images are divided into different human tissue regions according to different HU value ranges.

[0037] The conversion formula between pixel values ​​and HU values ​​in CT images is: HU = pixel * slope + intercept. Specifically, by reading two DICOM tag information from the DICOM format CT image: rescale intercept and rescale slope; and combining them with the aforementioned formula, the sum of the slope and intercept of the corresponding pixel is calculated to convert it into the corresponding HU value.

[0038] 3) Input the real human tissue distribution map and the predicted human tissue distribution map into the discriminator network of the first generative adversarial network, and improve the generation capability of the generator network of the first generative adversarial network based on adversarial training, so that the predicted human tissue distribution map output by it is more reasonable.

[0039] It should be noted that the predicted human tissue distribution map is used as a fake sample, and the real human tissue distribution map is used as a real sample. The discriminator network of the first generative adversarial network is trained with consistency constraints based on the adversarial principle to make the first generative adversarial network output a more reasonable predicted human tissue distribution map.

[0040] Step S2: Input the predicted human tissue distribution map and the non-attenuation corrected PET image into a second generative adversarial network cascaded with the first generative adversarial network to output the corresponding virtual CT image.

[0041] In one embodiment of this application, step S2 specifically includes:

[0042] a) Input the non-attenuation corrected PET image and the human tissue distribution prediction map into the generator network of the second generative adversarial network to obtain a CT prediction image with rich skeletal structure;

[0043] It should be noted that the generator network of the second generative adversarial network is trained to output a CT prediction image with rich skeletal structure by taking the obtained non-attenuation corrected PET image and the human tissue distribution prediction map output by the first generative adversarial network as input and the obtained real CT image as output.

[0044] b) Input the real CT image and the CT prediction image with rich skeletal structure into the discriminator network of the second generative adversarial network, and train the generator network of the second generative adversarial network based on the adversarial principle to output a virtual CT image.

[0045] It should be noted that the CT prediction image with rich skeletal structure output by the generator network of the second generative adversarial network is used as a fake sample, and the obtained real CT image is used as a real sample. Consistency constraint training is performed on the discriminator network of the second generative adversarial network so that the second generative adversarial network finally outputs a more realistic and reasonable virtual CT image for PET attenuation correction.

[0046] Step S3: Use the virtual CT image to perform attenuation correction on the non-attenuation-corrected PET image to obtain an attenuation-corrected PET image.

[0047] like Figure 2 The diagram illustrates a flowchart of PET image attenuation correction based on a human bone tissue enhancement framework according to an embodiment of this application. In the diagram, the NAC-PET image represents the non-attenuation-corrected PET image; the AC-PET image represents the attenuation-corrected PET image.

[0048] It should be noted that this application provides a bone enhanced generative framework (BEGF), comprising a cascaded first generative adversarial network (G1) and a second generative adversarial network (G2); as Figure 2 As shown in the dashed box (b), the bone tissue enhancement generation framework can generate virtual CT images for PET image attenuation correction. Compared to the existing PET attenuation correction framework (a) which only uses a single generative adversarial network, the bone tissue enhancement generation framework first constructs a human tissue distribution prediction map through a first generative adversarial network, then renders the CT volume through a second generative adversarial network, and finally outputs a virtual CT image with better tissue consistency and intensity similarity.

[0049] Specifically, firstly, an NAC-PET image is acquired and input into a first generative adversarial network (G1) to construct a human tissue distribution image, and outputs a more reasonable human tissue distribution prediction map; secondly, the more reasonable human tissue distribution prediction map output by the G1 network and the acquired NAC-PET image are input into a second generative adversarial network (G2) to render the CT volume, and output the corresponding virtual CT image; finally, the virtual CT image is used to perform attenuation correction on the NAC-PET image to obtain an AC-PET image.

[0050] In one embodiment of this application, the first generative adversarial network uses the Dice loss function for segment consistency constraints; the second generative adversarial network uses the mean absolute error loss function for projection consistency constraints.

[0051] It should be noted that the first generative adversarial network constrains the semantic information in the predicted human tissue distribution map through segmentation consistency constraints. The second generative adversarial network can also use the Dice loss function to perform projection consistency constraints to constrain the virtual CT image synthesized in the projection domain.

[0052] Specifically, the second generative adversarial network uses the mean absolute error loss function for projection consistency constraints, including: performing Radon transform on the virtual CT image and the acquired real CT image respectively to obtain the corresponding virtual CT projection and real CT projection; and applying the mean absolute error loss function to constrain the consistency between the virtual CT projection and the real CT projection. Projection consistency constraints improve the quality of the synthesized virtual CT image.

[0053] It should be noted that the Radon transform refers to projecting a digital image matrix along a ray direction at a specified angle. The integral operation of the Radon transform cancels out brightness fluctuations caused by noise. The Radon transform can be used to transform CT images from the image domain to the projection domain.

[0054] like Figure 3 The diagram shown illustrates the structure of a first generative adversarial network and a second generative adversarial network in one embodiment of this application.

[0055] In one embodiment of this application, the first generative adversarial network and the second generative adversarial network respectively include:

[0056] Generator Networks (G) x It is constructed using an encoder-decoder network with skip connections, consisting of 4 convolutional layers, 6 residual layers, and 2 deconvolutional layers;

[0057] Discriminator Network (D) x It includes 5 convolutional layers;

[0058] Both the convolutional layer and the deconvolutional layer employ 3D convolutional kernels.

[0059] Specifically, the generator network comprises: a single convolutional layer with a kernel size of 7×7×7, two convolutional layers with a kernel size of 3×3×3, six residual layers (RB), two deconvolutional layers with a kernel size of 3×3×3, and a single convolutional layer with a kernel size of 7×7×7, connected sequentially. Each residual layer consists of two convolutional layers with a kernel size of 3×3×3. The number of convolutional kernels in the generator network is 8, 16, 32, 32 (2×6), 32, 16, and 1, respectively; the convolutional stride is 1, 2, 2, 1, 1, 1, 1, 1, 1, 2, 2, 1, respectively.

[0060] In addition, residual layers are used to increase network depth and prevent the loss of features from human tissue distribution images and NAC-PET images during convolution; while skip connections are used to accelerate network training and preserve more image details. The deconvolution layer, also known as deconvolution, primarily serves an activation function, enabling the location of target human tissue in the image.

[0061] The discriminator contains 5 convolutional layers, all of which use 4×4×4 kernels; the number of kernels is 16, 32, 64, 128, and 1 respectively; and the convolution stride is 2, 2, 2, 2, and 1 respectively.

[0062] It is worth noting that the three most advanced existing methods for generating CT images from PET images include: 1) U-Net: directly learning the mapping from NAC-PAT images to CT images using U-Net; 2) CGAN: using U-Net as the backbone, with cyclic consistency loss and adversarial loss as additional constraints; and 3) AGAN: constraining CT image generation by performing segmentation sub-tasks on the generated CT images. A quantitative and qualitative comparison is performed between this method and the bone tissue enhancement-based generative framework BEGF provided in this application, using segmentation consistency and projection consistency as constraints, as follows:

[0063] A. Quantitative comparison

[0064] To enable quantitative comparison, we used intensity similarity and tissue consistency to measure the differences between the generated and target CT images. Intensity similarity was measured using four metrics: Peak Signal-to-Noise Ratio (PSNR), Mean Absolute Error (MAE), Normalized Cross-Correlation (NCC), and Structural Similarity Index (SSIM). Dice coefficients (denoted as Dice) were used for the segmented lung, liquid fat, soft tissue, and bone regions. a Dice c Dice t Dice b This is used to measure organizational consistency. The results are shown in Table 1.

[0065] From the results in the table, we can draw four conclusions: First, the CT images generated by U-Net have better intensity similarity but poor tissue consistency; this indicates that only MSE loss can maintain good intensity deviation, but it cannot distinguish different tissues and is insufficient for generating CT images. Second, CGAN achieves better tissue consistency than U-Net, but worse than AGAN and BEGF; this means that perceptual feature matching loss can implicitly identify different tissues, but it cannot be compared with explicit methods. Third, both AGAN and BEGF suggest that using explicit strategies to constrain tissue consistency can also achieve intensity similarity similar to CGAN and U-Net; this verifies the possibility that intensity similarity and tissue consistency can coexist well under appropriate constraints. Fourth, our proposed BEGF enhances tissue differences by constructing a skeleton separately, ultimately achieving the best results in both intensity similarity and tissue consistency for virtual CT images used for PET image attenuation correction. The results show that the bone tissue enhancement generation framework provided in this application is reasonable, and the strategy of enhancing tissue differences by constructing a skeleton separately is a reasonable and effective direction for NAC-PET to generate CT images.

[0066] Table 1. Quantitative comparison results of virtual CT images generated by different methods

[0067]

[0068] In this diagram, "↑" indicates that a higher value results in a better synthesized virtual CT image, while "↓" indicates that a lower value results in a better synthesized virtual CT image. For example, higher values ​​for Peak Signal-to-Noise Ratio (PANR), Normalized Cross-Correlation (NCC), and Structural Similarity Index (SSIM) indicate a better synthesized virtual CT image, while lower values ​​for Mean Absolute Error (MAE) indicate a better synthesized virtual CT image.

[0069] B. Qualitative Comparison

[0070] like Figure 4 The diagram illustrates a qualitative comparison of virtual CT images generated using different methods in one embodiment of this application. From left to right, the corresponding columns represent NAC-PET images, real CT images, and virtual CT images generated by U-Net, CGAN, AGAN, and BEGF of this application, respectively.

[0071] It should be noted that, for better visualization, we adjusted the contrast to [-400HU, 400HU]. From the figures, we can observe that because the lungs and air are areas with much lower intensity than other regions, they are easily identifiable in CT images generated by all methods. Conversely, skeletal regions are almost undetectable in CT images generated by U-Net and CGAN. Compared to other methods, our method generates CT images with significantly richer skeletal structures.

[0072] In summary, the bone enhancement generation framework BEGF provided in this application enhances tissue differences by constructing a skeleton separately, ultimately resulting in a virtual CT image for PET image attenuation correction. It not only achieves the best results in terms of intensity similarity and tissue consistency, but also has a richer skeletal structure.

[0073] like Figure 5 The diagram shown is a schematic representation of a PET image attenuation correction device according to an embodiment of this application.

[0074] The PET image attenuation correction device 500 includes:

[0075] The tissue prediction module 510 is used to acquire non-attenuation corrected PET images and input them into the first generative adversarial network to output a predicted map of human tissue distribution.

[0076] The virtual training module 520 is used to input the predicted human tissue distribution map and the non-attenuation corrected PET image into a second generative adversarial network cascaded with the first generative adversarial network, so as to output the corresponding virtual CT image.

[0077] The attenuation correction module 530 is used to perform attenuation correction on the non-attenuation-corrected PET image using the virtual CT image to obtain an attenuation-corrected PET image.

[0078] It should be understood that the division of the various modules in the above device is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can be implemented entirely in software via processing element calls; they can be fully implemented in hardware; or some modules can be implemented in software via processing element calls, while others are implemented in hardware. For example, the attenuation correction module 530 can be a separate processing element, or it can be integrated into a chip in the above device. Alternatively, it can be stored as program code in the memory of the above device, and its function can be called and executed by a processing element of the above device. The implementation of other modules is similar. Moreover, these modules can be fully or partially integrated together, or they can be implemented independently. 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 through integrated logic circuits in the hardware of the processor element or through software instructions.

[0079] For example, these modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more digital signal processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). As another example, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together to form a system-on-a-chip (SOC).

[0080] like Figure 6 The diagram shown illustrates the structure of a computer device 600 according to an embodiment of this application. The computer device 600 includes a memory 610 and a processor 620; the memory 610 stores computer instructions; the processor 620 executes the computer instructions to implement... Figure 1 The method described.

[0081] In some embodiments, the number of the memory 610 and the processor 620 in the computer device 600 can be one or more, while Figure 6 Each example is taken as an instance.

[0082] In one embodiment of this application, the processor 620 in the computer device 600 will perform as follows: Figure 1 The steps described involve loading one or more instructions corresponding to the process of an application into memory 610, and then having processor 620 run the application stored in memory 610, thereby achieving the following: Figure 1 The method described.

[0083] The memory 610 may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. The memory 610 stores an operating system and operating instructions, executable modules or data structures, or subsets thereof, or extended sets thereof. The operating instructions may include various operation instructions used to implement various operations. The operating system may include various system programs used to implement various basic business processes and handle hardware-based tasks.

[0084] The processor 620 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can 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, or discrete hardware components.

[0085] In some specific applications, the various components of the computer device 600 are coupled together through a bus system, which may include, in addition to a data bus, a power bus, a control bus, and a status signal bus, etc. However, for clarity, in... Figure 6 All kinds of buses are referred to as bus systems.

[0086] In one embodiment of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the following: Figure 1 The method described.

[0087] At any possible level of technical detail, this application can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this application.

[0088] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example, (but not limited to) electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0089] The computer-readable program described herein can be downloaded from a computer-readable storage medium to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives computer-readable program instructions from the network and forwards those instructions to the computer-readable storage medium stored in the respective computing / processing device.

[0090] The computer program instructions used to perform the operations of this application may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status 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, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via 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., via the Internet using an Internet service provider). In some embodiments, electronic circuits, such as programmable logic circuits, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), are personalized by utilizing the status information of the computer-readable program instructions. These electronic circuits can execute the computer-readable program instructions to implement various aspects of this application.

[0091] In summary, this application provides a PET image attenuation correction method, apparatus, and device, comprising: acquiring an unattenuation-corrected PET image and inputting it into a first generative adversarial network to output a predicted human tissue distribution map; inputting the predicted human tissue distribution map and the unattenuation-corrected PET image into a second generative adversarial network cascaded with the first generative adversarial network to output a corresponding virtual CT image; and using the virtual CT image to perform attenuation correction on the unattenuation-corrected PET image to obtain an attenuation-corrected PET image.

[0092] This application proposes a PET image attenuation correction method, apparatus, and device. The human bone tissue enhancement generative framework (BEGF) constructed based on generative adversarial network (GAN) has significantly richer bone structure in CT images generated from NAC-PET images. Furthermore, the CT images generated based on the BEGF network constructed in this application achieve better results in terms of intensity similarity and tissue consistency, thereby improving the accuracy of PET image attenuation correction.

[0093] This application effectively overcomes the various shortcomings of the prior art and has high industrial application value.

[0094] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for attenuation correction of PET images, characterized in that, The method includes: The process involves: acquiring unattenuated PET images and inputting them into a first generative adversarial network (GAN) to output a predicted human tissue distribution map. This includes: acquiring unattenuated PET images and inputting them into the generator network of the first GAN to obtain the predicted human tissue distribution map; acquiring real CT images, filtering them, and performing threshold segmentation based on pixel values ​​to obtain a real human tissue distribution map; inputting the real human tissue distribution map and the predicted human tissue distribution map into the discriminator network of the first GAN, and improving the generation capability of the generator network based on adversarial training to make the output predicted human tissue distribution map more reasonable; and performing threshold segmentation on the real CT images based on pixel values, including: defining regions in the CT images with CT values ​​in the range of [-1000 HU, -125 HU) as lung regions; defining regions in the CT images with CT values ​​in the range of [-125 HU, 10 HU) as fat regions; defining regions in the CT images with CT values ​​in the range of [10 HU, 90 HU) as soft tissue regions; and defining regions in the CT images with CT values ​​in the range of [90 HU, 1300 HU) as bone regions. The process involves inputting the predicted human tissue distribution map and the unattenuated PET image into a second generative adversarial network cascaded with the first generative adversarial network to output a corresponding virtual CT image. This includes: inputting the unattenuated PET image and the predicted human tissue distribution map into the generator network of the second generative adversarial network to obtain a CT prediction image with rich skeletal structure; inputting the real CT image and the CT prediction image with rich skeletal structure into the discriminator network of the second generative adversarial network, and training the generator network of the second generative adversarial network based on adversarial principles to output a virtual CT image. The virtual CT image is used to perform attenuation correction on the non-attenuation-corrected PET image to obtain an attenuation-corrected PET image.

2. The PET image attenuation correction method according to claim 1, characterized in that, The first generative adversarial network and the second generative adversarial network each include: The generator network is constructed using an encoder-decoder network with skip connections, consisting of 4 convolutional layers, 6 residual layers, and 2 deconvolutional layers. The discriminator network consists of 5 convolutional layers; Both the convolutional layer and the deconvolutional layer employ 3D convolutional kernels.

3. The PET image attenuation correction method according to claim 1, characterized in that, The first generative adversarial network uses the Dice loss function for segment consistency constraints; the second generative adversarial network uses the mean absolute error loss function for projection consistency constraints.

4. The PET image attenuation correction method according to claim 3, characterized in that, The second generative adversarial network uses the mean absolute error loss function for projection consistency constraints, including: The virtual CT image and the acquired real CT image are respectively subjected to Radon transform to obtain the corresponding virtual CT projection and real CT projection; The average absolute error loss function is used to apply consistent constraints to the virtual CT projection and the real CT projection.

5. A PET image attenuation correction device, characterized in that, include: The tissue prediction module is used to acquire unattenuated PET images and input them into a first generative adversarial network (GAN) to output a predicted human tissue distribution map. The module includes: acquiring unattenuated PET images and inputting them into the generator network of the first GAN to obtain the predicted human tissue distribution map; acquiring real CT images, filtering them, and performing threshold segmentation based on pixel values ​​to obtain a real human tissue distribution map; inputting the real human tissue distribution map and the predicted human tissue distribution map into the discriminator network of the first GAN, and improving the performance of the first GAN through adversarial training. The generative capability of the network is improved to make the predicted human tissue distribution map it outputs more reasonable; threshold segmentation of real CT images based on pixel values ​​includes: setting regions in CT images with CT values ​​in the range of [-1000HU, -125HU) as lung regions; setting regions in CT images with CT values ​​in the range of [-125HU, 10HU) as fat regions; setting regions in CT images with CT values ​​in the range of [10HU, 90HU) as soft tissue regions; and setting regions in CT images with CT values ​​in the range of [90HU, 1300HU) as bone regions. A virtual training module is used to input the predicted human tissue distribution map and the unattenuated PET image into a second generative adversarial network cascaded with the first generative adversarial network to output a corresponding virtual CT image. This module includes: inputting the unattenuated PET image and the predicted human tissue distribution map into the generator network of the second generative adversarial network to obtain a CT prediction image with rich skeletal structure; inputting the real CT image and the CT prediction image with rich skeletal structure into the discriminator network of the second generative adversarial network, and training the generator network of the second generative adversarial network based on adversarial principles to output a virtual CT image. The attenuation correction module is used to perform attenuation correction on the non-attenuation-corrected PET image using the virtual CT image to obtain an attenuation-corrected PET image.

6. A computer device, characterized in that, The device includes: a memory and a processor; The memory is used to store a computer program; the processor is used to execute the computer program stored in the memory to cause the device to perform the method as described in any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the PET image attenuation correction method according to any one of claims 1 to 4.

Images