Dual low dose attenuation correction method, system for pet / ct systems

Abstract

The scheme relates to a double low-dose attenuation correction method for a PET / CT system, belongs to the field of medical positron emission tomography, and is used for ensuring image quality while reducing radiation dose. The technical scheme is as follows: multi-scale feature extraction is performed on a low-dose ACCT image to obtain a first feature map set; multi-scale feature extraction is performed on a low-dose PET image without attenuation correction to obtain a second feature map set; N is a set value; the first feature map and the second feature map are adaptively aligned in space position, and the two aligned feature maps are matched and fused to obtain a third feature map set; M times of scale invariant feature extraction is performed on the third feature map to obtain an attenuation-corrected feature map, and M is a set value; the feature map is enlarged to have a size same as that of the feature map, and the feature map and the feature map are spliced to obtain an attenuation-corrected feature map; and a standard-dose PET image is obtained based on the feature map.

Description

Technical Field

[0001] This case relates to medical positron emission tomography, and more particularly to a dual low-dose attenuation correction method and system for PET / CT systems. Background Technology

[0002] Positron emission tomography combined with computed tomography (PET / CT) is a widely used medical system in the management of cancer patients, including diagnosis, monitoring, and follow-up treatment.

[0003] In PET imaging, attenuation correction improves visual interpretation and enables accurate quantitative analysis. CT-based methods are often used for PET attenuation correction, converting CT Hounsfield units to a linear attenuation coefficient at 511 keV energy through a bilinear relationship. However, this leads to spatial misalignment between PET and CT and limitations in CT-based artifacts.

[0004] Furthermore, patients, especially children, may face increased radiation risks after sequential scanning with a PET / CT scanner due to the potential hazards of CT dose and overexposure. While next-generation whole-body PET / CT scanners offer unprecedented image quality and quantitative accuracy, with approximately 40 times greater sensitivity, patients still receive significantly more radiation exposure during a full-body scan compared to single-site examinations. Summary of the Invention

[0005] To address the aforementioned problems in existing technologies, this invention aims to propose a dual low-dose attenuation correction method for PET / CT systems. This method considers radiation sources in both PET and CT imaging and utilizes the proposed dual dose reduction strategy. By reducing the injection dose and tube current, it directly generates normal-dose AC PET images from low-dose PET and low-dose ACCT (CT images utilized for attenuation correction) images, thereby reducing the radiation risk of the PET / CT system. This method can be used with whole-body PET / CT equipment to obtain normal-dose whole-body PET images.

[0006] To achieve the above technical objectives, the technical solution of this case is as follows.

[0007] Firstly, this application proposes a dual low-dose attenuation correction method for PET / CT systems, the method comprising the following steps:

[0008] Multi-scale feature extraction was performed on low-dose ACCT images to obtain the first feature map set. Multi-scale feature extraction was performed on low-dose, unattenuated PET images to obtain a second set of feature maps. N is a set value;

[0009] The first feature map Second feature map Adaptive spatial alignment is performed, and the two aligned feature maps are matched and fused to obtain a third feature map set.

[0010] The third feature map Perform M-fold scale-invariant feature extraction to obtain attenuation-corrected feature maps. M is a set value;

[0011] feature map Enlarge it to the same size as Same size, and and The attenuation correction feature map is obtained by splicing. j = N, ..., 2;

[0012] Based on feature maps Acquire visual PET images at standard doses.

[0013] In the above technical solution, one implementation of adaptive spatial alignment includes the following steps:

[0014] Based on the first feature map Obtaining the second feature map First feature map of the same size j = 2, 3, ..., N, while for j = 1, a first feature map with the same size as the cropped, low-dose, unattenuated corrected PET image is obtained.

[0015] Based on feature maps Perform an affine transformation to obtain the affine transformation parameters (γ, β), j = 1, 2, ..., N;

[0016] Using affine transformation parameters (γ, β), the feature map is... Scaling and translation operations are performed on the channel to obtain the second feature map.

[0017] In the above technical solution, one implementation method for matching and fusion includes the following steps:

[0018] Based on feature maps and Matching is achieved using the Hadamard product operation to obtain feature maps.

[0019] Based on the first feature map Obtaining the second feature map First feature map of the same size j = 2, 3, ..., N, while for j = 1, a first feature map of the same size as the cropped, low-dose, unattenuated corrected PET image is obtained.

[0020] feature map and The fusion is achieved using the Hadamah operation, where j = 1, 2, ..., N.

[0021] Secondly, this case proposes a dual low-dose attenuation correction system for PET / CT systems. The system uses a dual low-dose attenuation calibration model to take low-dose ACCT images and low-dose uncorrected PET images as inputs, and standard-dose attenuation-corrected PET images as outputs.

[0022] The dual low-dose attenuation calibration model includes a first encoder, a second encoder, a spatial alignment module, and a decoder, wherein:

[0023] The first encoder is configured to perform multi-scale feature extraction on low-dose ACCT images to obtain a first feature map set. N is a set value;

[0024] The second encoder is configured to perform multi-scale feature extraction on low-dose, unattenuated PET images to obtain a second feature map set.

[0025] The spatial alignment module is configured to align the first feature map. Second feature map Adaptive spatial alignment is performed, and the two aligned feature maps are matched and fused to obtain a third feature map set.

[0026] The decoder is configured to implement feature maps Upsampling, making it consistent with Same size, and and The attenuation correction feature map is obtained by splicing. j = N, ..., 2; where: feature map Based on the third feature map The feature is obtained by performing M scale-invariant feature extractions, where M is a set value.

[0027] In the above technical solution, one implementation of the first encoder, the second encoder, and the decoder is that they are all composed of residual modules;

[0028] Each residual module that makes up the encoder is composed of two three-dimensional convolutional layers, with a batch normalization layer and an activation function layer placed between the two convolutional layers;

[0029] Each residual module that makes up the decoder consists of a three-dimensional transposed deconvolutional layer and a convolutional layer, with a batch normalization layer and an activation function layer between the transposed deconvolutional layer and the convolutional layer.

[0030] In one embodiment of the above technical solution, the loss function for training the dual low-dose attenuation calibration model is as follows:

[0031]

[0032] In the formula: G represents the mapping relationship, θ i Represents network parameters, x i For samples in the low-dose ACCT image dataset, y i For samples in a low-dose, unattenuated PET image dataset with multiple dose levels, z i represents the samples in the PET image dataset after standard dose attenuation correction, and n is the total number of training samples.

[0033] In one embodiment of the above technical solution, the spatial alignment module includes a clipping unit, an affine unit, and a scaling and translation unit; wherein:

[0034] The cropping unit is configured based on the first feature map. Obtaining the second feature map First feature map of the same size For j = 2, 3, ..., N, and for j = 1, a first feature map of the same size as the cropped, low-dose, unattenuated corrected PET image is obtained.

[0035] Affine units are configured to be based on feature maps. Perform an affine transformation to obtain the affine transformation parameters (γ, β), j = 1, 2, ..., N;

[0036] The scaling and translation unit is configured to apply an affine transformation parameter (γ, β) to the feature map. Scaling and translation operations are performed on the channel to obtain the second feature map.

[0037] In one embodiment of the above technical solution, the affine unit uses a filter and a sigmoid function to obtain the affine transformation parameters.

[0038] In one embodiment of the above technical solution, the spatial alignment module further includes a matching unit and a fusion unit; wherein:

[0039] The matching unit is configured to be based on feature maps. and Matching is achieved using the Hadamard product operation to obtain feature maps.

[0040] The fusion unit is configured to fuse feature maps and Fusion is achieved using the Hadamard operation, where j = 1, 2, ..., N, and feature map By obtaining it again from the trimming unit.

[0041] In one embodiment of the above technical solution, M residual modules are used to process the third feature map. Perform M-fold scale-invariant feature extraction to obtain attenuation-corrected feature maps. M is a set value.

[0042] The beneficial technical effects of this case are:

[0043] Based on low-dose unattenuated PET images and low-dose ACCT images, attenuated PET images are directly generated using a deep learning model, significantly reducing the radiation risk to patients, especially beneficial for the diagnosis, monitoring, and follow-up treatment of pediatric diseases. When used for whole-body imaging, it reduces CT scan time and minimizes respiratory artifacts during CT scans, thus ensuring the quality of attenuated PET images. Therefore, this approach ensures both image quality and reduces harmful radiation dose, possessing significant scientific value and promising application prospects in the field of medical diagnostics. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 , one A schematic diagram of the overall system framework in one implementation method;

[0046] Figure 2 , one A schematic diagram illustrating image size changes during spatial alignment in one implementation method;

[0047] Figure 3 , one Attenuation-corrected PET images in one implementation method;

[0048] Figure 4 , one A schematic diagram of a reference standard image in one embodiment;

[0049] Figure 5 , one The PET image without attenuation correction in one embodiment. Detailed Implementation

[0050] PET attenuation correction requires CT images to calculate the attenuation correction coefficient map, resulting in patients receiving a high radiation dose. Most current methods for low-radiation-dose whole-body PET / CT imaging ignore the actual physical characteristics of the equipment and the scanning protocol, or in other words, do not take into account the dose of ACCT (CT images utilized for attenuation correction).

[0051] Therefore, this solution considers the radiation sources in PET and CT imaging. Utilizing a deep learning model, it directly generates normal-dose ACPET images from low-dose PET and low-dose ACCT images while meeting clinical application requirements, thereby reducing the radiation risk of the PET / CT system. This invention can be used for both local (e.g., brain) PET attenuation correction and whole-body PET attenuation correction without concerns about increased radiation. Furthermore, the dual low-dose strategy shortens CT scan time, thus suppressing respiratory artifacts during whole-body CT scans, improving the quality of PET attenuation correction, and enhancing the accuracy of quantitative analysis of PET images.

[0052] The following description, in conjunction with the accompanying drawings, clearly and completely describes how the technical solution of this case is implemented. Obviously, the described embodiments are only a part of the embodiments of this case, and not all of them. Based on the embodiments in this case, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this application.

[0053] Figure 1 This is a schematic diagram of the dual low-dose attenuation calibration model used when implementing a dual low-dose attenuation correction system for PET / CT systems.

[0054] The dual low-dose attenuation calibration model includes a first encoder (first gray column), a second encoder (second green column), a spatial alignment module, and a decoder (third blue column). See also Figure 1Before extracting features from the image, the first and second encoders first expand the number of channels in the image to increase its dimensionality. Correspondingly, at the end of the decoder, the number of channels is restored to obtain the visible image. In the feature map processing section, for example, the first encoder, second encoder, and decoder each consist of five residual modules.

[0055] The first encoder is used to extract features from the input low-dose ACCT image. Each residual module is implemented using two 3D convolutional layers (using a 3×3×3 filter size). A batch normalization layer and an activation function layer are added between these two convolutional layers, which can be represented as:

[0056] f j+1 =F j (f j )=g(w j f j +b j )

[0057] In the formula: f j+1 w represents the feature map obtained after passing through the residual module. j and b j This represents the weights and biases corresponding to the convolution operation.

[0058] The encoder output for low-dose ACCT images can be expressed as:

[0059] f CT =F L (…F1(x i ))

[0060] In the formula: f CT x represents the ACCT image output feature map. i This represents a low-dose ACCT image sample, F j L represents the number of residual modules, thus obtaining multi-scale features, which are denoted as the first feature map set. See the diagram illustrating image size changes. Figure 2 Corresponding column.

[0061] Similarly, for low-dose PET images, a second encoder with the same design is introduced to extract features from the input low-dose NAC PET image, which can also be represented as:

[0062] f PET =F L (…F1(y i ))

[0063] In the formula: f CT =F L (…F1(x i)) represents the output feature map of a low-dose, uncorrected PET image, y i This indicates that low-dose, unattenuated PET image samples, after feature extraction by the residual module, are downsampled using a 3×3×3 convolution with a stride of 2, allowing the network to obtain features at different scales. The resulting multi-scale features are denoted as... See the diagram illustrating image size changes. Figure 2 Corresponding column.

[0064] The input low-dose ACCT images (512×512) and NAC PET images (192×192) have size differences. Therefore, a spatial alignment module is designed to facilitate image feature matching and fusion at different scales for the features of the two modalities.

[0065] First, the NAC (Non-attenuation-corrected) PET image is cropped to a size of 128×128. Then, the first feature map corresponding to the ACCT image is processed. Two downsampling operations are performed using a 3×3×3 convolutional kernel with a stride of 2, resulting in a more efficient feature map. and feature images Same size, j = 2, 3, ..., N; for j = 1, obtain a first feature map with the same size as the cropped, low-dose, unattenuated corrected PET image.

[0066] Based on feature maps Perform an affine transformation to obtain the affine transformation parameters (γ, β), j = 1, 2, ..., N:

[0067]

[0068] In the formula: H represents affine transformation.

[0069] In one specific implementation, affine transformation parameters (γ, β) are obtained from the intermediate feature map of ACCT using a 3×3×3 filter and a sigmoid function, which are then used to refine the feature image of the PET image. Scaling and translation operations on a channel can be specifically represented as follows:

[0070]

[0071] In the aforementioned spatial alignment module, adaptive affine transformation of the channel dimension of the intermediate feature map is performed to promote feature matching and fusion of low-dose PET and ACCT images.

[0072] Figure 1The lower left section illustrates a matching and fusion method. Specifically, it is based on feature maps. and Matching is achieved using the Hadamard product operation to obtain feature maps. Based on the first feature map Obtaining the second feature map First feature map of the same size j = 2, 3, ..., N, while for j = 1, a first feature map of the same size as the cropped, low-dose, unattenuated corrected PET image is obtained. feature map and The fusion is achieved using the Hadamard operation, where j = 1, 2, ..., N. During this process, the feature maps are... and During matching, downsampling was performed twice; after matching, data was re-sampled from the clipping unit. That is, to re-apply the feature map After downsampling, the results are fused with the matching results to obtain the output of the spatial alignment module, which is denoted as the third feature map set.

[0073] For the third feature map Perform M-fold scale-invariant feature extraction to obtain attenuation-corrected feature maps. M is a set value. For example, three cascaded residual modules are used to achieve three-fold scale-invariant feature extraction.

[0074] Will The input is to the decoder, which predicts normal-dose AC-PET images through upsampled transposed convolutional operations and short connection operations. The decoder also has five residual modules, each implemented using a 3D transposed deconvolutional layer (with a 3×3×3 filter size and a stride of 2) and a convolutional layer (with a 3×3×3 filter size and a stride of 1). Batch normalization layers and activation function layers are added between these two transposed deconvolutional layers. Skip connection operations connect the fused PET and ACCT image features with upsampled features at the corresponding scale, preserving more contextual information.

[0075] See Figure 1 The blue column, from bottom to top, shows the feature maps processed by the decoder. Upsample and amplify to make its size the same as... Same size, and and Attenuation correction feature map is obtained by splicing through skip connections. j = N, ..., 2. See the diagram illustrating image size changes. Figure 2 Corresponding columns. Obtain the feature map. Then, a visual, standard measurement PET image is obtained by channel reconstruction.

[0076] In summary, the dual low-dose attenuation correction system for PET / CT systems uses a dual low-dose attenuation calibration model to take low-dose ACCT images and low-dose uncorrected PET images as inputs, and standard-dose attenuation-corrected PET images as outputs.

[0077] Attenuation correction is a commonly used technique in nuclear medicine imaging (such as positron emission tomography, PET-CT) to correct for changes in image brightness caused by the attenuation of radioactive isotopes. It compensates for image degradation, making radiation absorption in different regions more accurate and reliable. This study trained a dual low-dose attenuation correction model to correct low-dose uncorrected PET images using low-dose ACCT images, resulting in attenuation-corrected PET images with standard doses. The training dataset (PET and CT data) was obtained from a uEXPLORER whole-body PET / CT scanner with a tube voltage of 120 kVp and a tube current of 20 mA. The dataset included uncorrected whole-body PET images at various dose levels. To simulate noise distribution in PET images, noise was roughly divided into four categories, denoted as 1.0%, 2.5%, 5.0%, and 25% doses. All simulated doses were extracted from the standard raw data by randomly selecting a certain proportion of count events. The reconstructed PET image size was 192 × 192 × 673 pixels, with a thickness of 2.9 mm. For low-dose ACCT images, the scanning tube voltage was 120 kVp and the tube current was 20 mA. The thickness of the low-dose ACCT image was set to 3 mm. The resulting ACCT image size was 512 × 512 × 673.

[0078] Suppose we have a training dataset D = {(x1, y1), (x2, y2), ..., (x...} n y n )}, where {x1, x2, ..., x n} represents samples from the low-dose ACCT image dataset, {y1, y2, ..., y} n} is a sample from a low-dose, unattenuated corrected PET image dataset, {z1, z2, ..., z} n} represents the samples in the PET image dataset after standard dose attenuation correction, and n is the total number of training samples.

[0079] Next, a loss function is constructed for training the low-dose attenuation calibration model. One implementation uses the MSE function as the loss function, and the training loss can be expressed as:

[0080]

[0081] In the formula: G represents the mapping relationship, θ i Represents network parameters, x i For samples in the low-dose ACCT image dataset, y i For samples in a low-dose, unattenuated PET image dataset with multiple dose levels, z i represents the samples in the PET image dataset after standard dose attenuation correction, and n is the total number of training samples.

[0082] The dual low-dose attenuation calibration model was trained using the aforementioned training dataset, i.e., low-dose, unattenuated corrected whole-body PET images. i (see Figure 3 (as shown) and low-dose ACCT images x i (see Figure 4 (As shown) PET image z after attenuation correction at standard dose, used as network input. i As a reference standard (ground truth) (see...) Figure 5 As shown in the figure, the Adam optimizer is used for optimization.

[0083] The trained dual low-dose attenuation calibration model processes low-dose ACCT images and low-dose uncorrected PET images as follows:

[0084] Multi-scale feature extraction was performed on low-dose ACCT images to obtain the first feature map set. Multi-scale feature extraction was performed on low-dose, unattenuated PET images to obtain a second set of feature maps. N is a set value;

[0085] The first feature map Second feature map Adaptive spatial alignment is performed, and the two aligned feature maps are matched and fused to obtain a third feature map set.

[0086] The third feature map Perform M-fold scale-invariant feature extraction to obtain attenuation-corrected feature maps. M is a set value;

[0087] feature map Enlarge it to the same size as Same size, and and The attenuation correction feature map is obtained by splicing. j = N, ..., 2;

[0088] Based on feature maps Acquire visual PET images at standard doses.

[0089] As can be seen from the above implementation process, the dual dose reduction strategy employed in this case significantly lowers the radiation dose in whole-body PET / CT scans. This is of great significance for ultra-low-dose scanning protocols for whole-body scans, ensuring the integrity of the scanning protocol. When performing whole-body PET attenuation correction, reducing the ACCT dose rather than eliminating it directly makes it more clinically acceptable. Currently, whole-body CT scans take approximately 20 seconds and introduce respiratory artifacts. However, the dual low-dose whole-body PET attenuation correction strategy reduces CT scan time and minimizes respiratory artifacts present during CT scans, thus ensuring the quality of the attenuated PET images. This case demonstrates the acquisition of PET images at the lowest clinically acceptable dose level while simultaneously reducing both PET and CT doses.

[0090] It should be noted that this technical solution can be applied to tracer PET / SPECT imaging, which can help provide diagnostic references and develop clinical diagnostic guidelines.

[0091] In the above process, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first," "second," or "third" may explicitly or implicitly include one or more of that feature.

[0092] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods or systems disclosed herein can be implemented using software plus necessary general-purpose hardware, or they can be implemented using dedicated hardware including dedicated integrated circuits, dedicated CPUs, dedicated memory, dedicated components, etc. Generally, any function performed by a computer program can be easily implemented using corresponding hardware, and the specific hardware structure used to implement the same function can be diverse, such as analog circuits, digital circuits, or dedicated circuits. However, for the purposes of this disclosure, software program implementation is more often a preferred implementation method.

[0093] It should be noted that the terms "one embodiment," "another embodiment," and "embodiment" used in this specification refer to specific features, structures, or characteristics described in connection with that embodiment, which are included in at least one embodiment described in the general description of this application. The appearance of the same expression in multiple places in the specification does not necessarily refer to the same embodiment. Furthermore, when a specific feature, structure, or characteristic is described in connection with any embodiment, the intention is to suggest that implementing such a feature, structure, or characteristic in conjunction with other embodiments also falls within the scope of this invention.

[0094] Although embodiments of the present invention have been described above in conjunction with the accompanying drawings, the present invention is not limited to the specific embodiments and application fields described above. The specific embodiments described above are merely illustrative and instructive, and not restrictive. Those skilled in the art can make many other forms based on the guidance of this specification and without departing from the scope of protection of the claims of the present invention, and all of these are within the scope of protection of the present invention.

Claims

1. A dual low-dose attenuation correction method for PET / CT systems, characterized in that, The method includes the following steps: Multi-scale feature extraction was performed on low-dose ACCT images to obtain the first feature map set. Multi-scale feature extraction was performed on low-dose, unattenuated PET images to obtain a second feature map set. N is a set value; The first feature map Second feature map Adaptive spatial alignment is performed, and the two aligned feature maps are matched and fused to obtain a third feature map set. ; The third feature map Perform M-fold scale-invariant feature extraction to obtain attenuation-corrected feature maps. M is a set value; feature map Enlarge it to the same size as Same size, and and The attenuation correction feature map is obtained by stitching the images together. , ; Based on feature maps Acquire visual PET images of standard doses; The adaptive spatial alignment steps include: Based on the first feature map Obtain the second feature map First feature map of the same size For j=2,3,…,N, and for j=1, a first feature map of the same size as the cropped, low-dose, unattenuated corrected PET image is obtained. ; Based on feature maps Perform an affine transformation to obtain the affine transformation parameters. , ; Using affine transformation parameters For feature maps Scaling and translation operations are performed on the channel to obtain the second feature map. , , .

2. The method according to claim 1, characterized in that, The matching and fusion process involves the following steps: Based on feature maps and Matching is achieved using the Hadamard product operation to obtain feature maps. ; Based on the first feature map Obtain the second feature map First feature map of the same size For j=2,3,…,N, and for j=1, a first feature map of the same size as the cropped, low-dose, unattenuated corrected PET image is obtained. ; feature map and Fusion is achieved using Hadamagga operations. .

3. A dual low-dose attenuation correction system for PET / CT systems, characterized in that: The system uses a dual low-dose attenuation calibration model to take low-dose ACCT images and low-dose uncorrected PET images as inputs, and standard-dose attenuation-corrected PET images as outputs. The dual low-dose attenuation calibration model includes a first encoder, a second encoder, a spatial alignment module, and a decoder, wherein: The first encoder is configured to perform multi-scale feature extraction on low-dose ACCT images to obtain a first feature map set. N is a set value; The second encoder is configured to perform multi-scale feature extraction on low-dose, unattenuated PET images to obtain a second feature map set. ; The spatial alignment module is configured to align the first feature map. Second feature map Adaptive spatial alignment is performed, and the two aligned feature maps are matched and fused to obtain a third feature map set. ; The decoder is configured to implement feature maps Upsampling, making it consistent with Same size, and and The attenuation correction feature map is obtained by stitching the images together. , Among them: feature map Based on the third feature map The feature is obtained by performing M scale-invariant feature extractions, where M is a set value. The adaptive spatial alignment steps include: Based on the first feature map Obtain the second feature map First feature map of the same size For j=2,3,…,N, and for j=1, a first feature map of the same size as the cropped, low-dose, unattenuated corrected PET image is obtained. ; Based on feature maps Perform an affine transformation to obtain the affine transformation parameters. , ; Using affine transformation parameters For feature maps Scaling and translation operations are performed on the channel to obtain the second feature map. , , .

4. The system according to claim 3, characterized in that: The first encoder, the second encoder, and the decoder are all composed of residual modules; Each residual module that makes up the encoder is composed of two three-dimensional convolutional layers, with a batch normalization layer and an activation function layer placed between the two convolutional layers; Each residual module that makes up the decoder consists of a three-dimensional transposed deconvolutional layer and a convolutional layer, with a batch normalization layer and an activation function layer between the transposed deconvolutional layer and the convolutional layer.

5. The system according to claim 3, characterized in that: The loss function for training the dual low-dose attenuation calibration model is as follows: In the formula: Indicates the mapping relationship. Represents network parameters, For samples from the low-dose ACCT image dataset, This is a sample from a dataset of low-dose, unattenuated PET images with multiple dose levels. represents the samples in the PET image dataset after standard dose attenuation correction, and n is the total number of training samples.

6. The system according to claim 3, characterized in that: The spatial alignment module includes clipping units, affine units, and scaling / translation units; among which: The cropping unit is configured based on the first feature map. Obtain the second feature map First feature map of the same size For j=2,3,…,N, and for j=1, a first feature map of the same size as the cropped, low-dose, unattenuated corrected PET image is obtained. ; Affine units are configured to be based on feature maps. Perform an affine transformation to obtain the affine transformation parameters. , ; The scaling and translation unit is configured to utilize affine transformation parameters. For feature maps Scaling and translation operations are performed on the channel to obtain the second feature map. , , .

7. The system according to claim 6, characterized in that: The affine unit uses filters and the sigmoid function to obtain the affine transformation parameters.

8. The system according to claim 6, characterized in that, The spatial alignment module also includes a matching unit and a fusion unit; wherein: The matching unit is configured to be based on feature maps. and Matching is achieved using the Hadamard product operation to obtain feature maps. ; The fusion unit is configured to fuse feature maps and Fusion is achieved using Hadamagga operations. , where: feature map By obtaining it again from the trimming unit.

9. The system according to claim 3, characterized in that: The third feature map is processed using M residual modules. Perform M-fold scale-invariant feature extraction to obtain attenuation-corrected feature maps. M is a set value.

Images