A PET attenuation correction method, system, device and storage medium

Abstract

The present invention discloses a PET attenuation correction method, system, device and storage medium. In an embodiment of the present invention, an image segmentation network is first used to perform image transformation processing on the image to be reconstructed, so as to realize image denoising, filtering and image dose mapping conversion, thereby providing a good image data basis for PET attenuation correction; then, a target projection tensor is obtained through back-projection processing to retain more features and provide more useful information for subsequent processing; further, group convolution is used to reduce the number of convolution parameters, thereby accelerating the convolution and PET attenuation correction speed; finally, global feature extraction of a residual dense network is used to avoid the influence of gradient vanishing or explosion caused by the increase of network depth, thereby achieving excellent reconstruction effect, and the present invention can be widely used in the field of image processing technology.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a PET attenuation correction method, system, device and storage medium. Background Art

[0002] Positron emission tomography (PET) is an important part of modern imaging technology in the field of nuclear medicine. Quantitative analysis of the metabolic process of tracers in the body is an important advantage of its imaging. The principle of PET imaging is that radioactive nuclides decay to release positrons, which annihilate radiation in human tissues and produce a pair of γ photons with opposite directions and equal energy (511keV). In the process of passing through an object to reach the detector, there is a certain probability that γ photons will undergo photoelectric effect or Compton scattering with electrons in the object, resulting in failure to be effectively detected by the detector. This phenomenon is called attenuation. Attenuation will seriously affect the accuracy of quantitative analysis of the distribution of radioactive drugs in the body and is the most serious factor affecting the quantitative accuracy of PET images. Therefore, attenuation correction (AC) is essential. Attenuation correction is mainly achieved by measuring the attenuation ratio of annihilation radiation photons (I0 / I) or estimating the μ value of each line response (LOR), and then applying it to image processing to compensate for the counting loss caused by attenuation, and finally obtain an image of the actual radioactivity distribution of the tissue.

[0003] Based on the characteristics of PET imaging, the transmission source correction method is often used. Compared with the early PET transmission scanning method using an external 68Ge rod source, PET / CT uses CT data to perform attenuation correction on PET images (CT-based Attenuation Correction, CTAC), which has the advantages of fast speed, good counting, and low noise. It is the gold standard of current PET attenuation correction technology. However, the dose problem caused by CTAC has also put forward new requirements for attenuation correction technology.

[0004] With the research and development of artificial intelligence in the field of medical imaging, many studies have proposed PET attenuation correction technology based on machine learning. As a new field in machine learning research, deep learning has made particularly significant progress in the research of PET attenuation correction technology in recent years. Deep learning uses a multi-layer neural network containing a large number of neurons to extract useful features from images. The research on attenuation correction mainly focuses on synthesizing MR images or PET images into pseudo-CT images or attenuation coefficient maps (μ-maps). However, such methods have errors caused by the misalignment and poor registration between different mode images when training and testing the network. In view of this, how to achieve PET attenuation correction with better generalization ability and higher accuracy is an urgent problem to be solved. Summary of the invention

[0005] In view of this, embodiments of the present invention provide a PET attenuation correction method, system, device and storage medium, which can efficiently implement PET attenuation correction.

[0006] In one aspect, an embodiment of the present invention provides a PET attenuation correction method, comprising:

[0007] Acquire PET data to be corrected; the PET data to be corrected includes continuous multi-layer PET images;

[0008] The PET data to be corrected is converted by using a diffusion model to obtain CT data; wherein the diffusion model is generated by training the PET data paired with the CT data; and the CT data includes continuous multi-layer CT images;

[0009] determining a grouping result of the CT data based on the marked boundary, and dividing the PET data to be corrected into first PET data and second PET data according to the grouping result;

[0010] Using the three-dimensional model, performing a first attenuation correction on the first PET data to obtain first corrected data; wherein the three-dimensional model is generated by training the PET data of the paired three-dimensional blocks;

[0011] Performing a second attenuation correction on the second PET data using the two-dimensional model to obtain second corrected data; wherein the two-dimensional model is generated by training the PET data of the paired two-dimensional sequence;

[0012] Target PET data is obtained according to the first correction data and the second correction data.

[0013] Optionally, using a diffusion model, data conversion is performed on the PET data to be corrected to obtain CT data, including:

[0014] Using the diffusion model, each PET image of the PET data to be corrected is subjected to forward diffusion processing to obtain Gaussian noise;

[0015] According to Gaussian noise, the CT image corresponding to the PET image is restored through reverse sampling processing, and the CT image is integrated to obtain CT data.

[0016] Optionally, the method further comprises the step of obtaining a three-dimensional model by training the PET data of the paired three-dimensional blocks, the step comprising:

[0017] Acquire PET data;

[0018] Performing three-dimensional block extraction on the PET data through a three-dimensional extraction component, and converting to obtain PET data in a three-dimensional block format;

[0019] Constructing a first residual network; the first residual network includes a first feature extraction module, a first feature encoding module, a first feature capture module and a first normalization module;

[0020] Using PET data in a three-dimensional block format as a training dataset, and using PET data in a paired three-dimensional block format as label data;

[0021] The first residual network is trained and optimized through the training data set and the label data, and based on the training result, the first residual network is adjusted to obtain a three-dimensional model.

[0022] Optionally, performing a first attenuation correction on the first PET data using the three-dimensional model to obtain first corrected data includes:

[0023] The first PET data is subjected to first feature extraction, first feature encoding, first feature capture and first normalization processing to obtain first corrected data.

[0024] Optionally, the method further comprises the step of obtaining a two-dimensional model by training the PET data of the paired two-dimensional sequence, and this step comprises:

[0025] Acquire PET data; wherein the PET data includes continuous multi-layer PET images, and the PET data is in a two-dimensional sequence format;

[0026] Constructing a second residual network; the second residual network includes a second feature extraction module, a second feature encoding module, a second feature capture module and a second normalization module;

[0027] Using PET data in two-dimensional sequence format as training data set, and using PET data in paired two-dimensional sequence format as label data;

[0028] The second residual network is trained and optimized through the training data set and the label data, and based on the training result, the second residual network is adjusted to obtain a two-dimensional model.

[0029] Optionally, performing a second attenuation correction on the second PET data using the two-dimensional model to obtain second corrected data includes:

[0030] The second PET data is subjected to second feature extraction, second feature encoding, second feature capture and second normalization processing to obtain second corrected data.

[0031] Optionally, obtaining target PET data according to the first correction data and the second correction data includes:

[0032] Based on the boundary corresponding to the grouping result, the first corrected data and the second corrected data are spliced ​​and fused to obtain the target PET data.

[0033] On the other hand, an embodiment of the present invention provides a PET attenuation correction system, comprising:

[0034] The first module is used to obtain PET data to be corrected; the PET data to be corrected includes continuous multi-layer PET images;

[0035] The second module is used to perform data conversion on the PET data to be corrected by using a diffusion model to obtain CT data; wherein the diffusion model is generated by training the PET data paired with the CT data; and the CT data includes continuous multi-layer CT images;

[0036] A third module is used to determine a grouping result of the CT data based on the marked boundary, and divide the PET data to be corrected into first PET data and second PET data according to the grouping result;

[0037] A fourth module is used to perform a first attenuation correction on the first PET data using a three-dimensional model to obtain first corrected data; wherein the three-dimensional model is generated by training the PET data of the paired three-dimensional blocks;

[0038] A fifth module is used to perform a second attenuation correction on the second PET data using the two-dimensional model to obtain second corrected data; wherein the two-dimensional model is generated by training the PET data of the paired two-dimensional sequence;

[0039] The sixth module is used to obtain target PET data according to the first correction data and the second correction data.

[0040] On the other hand, an embodiment of the present invention provides a PET attenuation correction device, including a processor and a memory;

[0041] The memory is used to store programs;

[0042] The processor executes the program to implement the above method.

[0043] On the other hand, an embodiment of the present invention provides a computer-readable storage medium, wherein the storage medium stores a program, and the program is executed by a processor to implement the above method.

[0044] The embodiment of the present invention also discloses a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device can read the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the above method.

[0045] The embodiment of the present invention first obtains PET data to be corrected; the PET data to be corrected includes continuous multi-layer PET images; using a diffusion model, the PET data to be corrected is converted to obtain CT data; wherein the diffusion model is generated by training PET data of paired CT data; the CT data includes continuous multi-layer CT images; based on the marked boundary, the grouping result of the CT data is determined, and the PET data to be corrected is divided into first PET data and second PET data according to the grouping result; using a three-dimensional model, a first attenuation correction is performed on the first PET data to obtain first corrected data; wherein the three-dimensional model is generated by training PET data of paired three-dimensional blocks; using a two-dimensional model, a second attenuation correction is performed on the second PET data to obtain second corrected data; wherein the two-dimensional model is generated by training PET data of paired two-dimensional sequences; according to the first corrected data and the second corrected data, the target PET data is obtained. The embodiment of the present invention first converts the PET image into a clearer CT image through data conversion, which is convenient for subsequent boundary division, and then adopts correction classification processing based on three-dimensional and two-dimensional based on the grouping result, and fully plans the data processing calculation degree to avoid calculation redundancy on the premise of ensuring the quality and accuracy of attenuation correction in a specific area. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1 A schematic diagram of a flow chart of a PET attenuation correction method provided by an embodiment of the present invention;

[0048] Figure 2 A data flow diagram of data conversion provided by an embodiment of the present invention;

[0049] Figure 3 A schematic diagram of the flow of training and application of a diffusion model provided by an embodiment of the present invention;

[0050] Figure 4 A schematic diagram of the structure of a residual network provided by an embodiment of the present invention;

[0051] Figure 5 A schematic diagram of the process of training and applying a three-dimensional model provided in an embodiment of the present invention;

[0052] Figure 6 A schematic diagram of the process of training and applying a two-dimensional model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0054] On the one hand, if Figure 1 As shown, an embodiment of the present invention provides a PET attenuation correction method, comprising:

[0055] S100, obtaining PET data to be corrected;

[0056] The PET data to be corrected include continuous multi-layer PET images;

[0057] S200, using a diffusion model, performing data conversion on the PET data to be corrected to obtain CT data;

[0058] The diffusion model is generated by training PET data paired with CT data; the CT data includes continuous multi-layer CT images; it should be noted that, in some embodiments, the diffusion model is used to perform forward diffusion processing on each PET image of the PET data to be corrected to obtain Gaussian noise; based on the Gaussian noise, the CT image corresponding to the PET image is restored by reverse sampling processing, and the CT image is integrated to obtain the CT data;

[0059] Specifically, in some specific embodiments, by inputting none-AC PET data (data without attenuation correction, i.e., PET data to be corrected) into a deep learning network model, the corresponding virtual CT data is output; wherein, PET data and CT data refer to PET images and CT images, which generally produce continuous multi-layer images, thereby reflecting the conditions of internal organ tissues at the irradiated site; the deep learning network model can be selected as some classic models, such as the diffusion model (DiffusionModel), which can realize the conversion of data of different modalities (input A modality data, after passing through the network model, output B modality data), and the model also needs to be trained in advance. Taking the diffusion model as an example, the training steps are as follows:

[0060] 1. Training data set: Input: none-AC PET image; Label: paired CT image;

[0061] 2. Perform model training optimization based on the training data set and labels, and make adjustments based on the training results to obtain the diffusion model; model output: synthetic CT image;

[0062] Data conversion is performed through the trained diffusion model, such as Figure 2As shown in the figure (X0 and Z in the figure represent the processing variables involved in the data processing process of the model, which are not restrictive), the diffusion model gradually adds noise to the image during the forward diffusion process until the image is destroyed and becomes completely Gaussian noise, and then learns to restore the real image from Gaussian noise during the reverse sampling process. Figure 3 As shown in Figure 2, through training, the diffusion model G from the whole-body none-AC PET image to the whole-body CT image is obtained. CT , the whole-body none-AC PET data obtained by scanning can be directly input into the trained model G CT , and obtain a synthetic whole-body CT image.

[0063] S300, determining a grouping result of the CT data based on the marked boundary, and dividing the PET data to be corrected into first PET data and second PET data according to the grouping result;

[0064] Among them, the first PET data represents a set of PET images of a highly heterogeneous area, and the PET images of the corresponding area usually require high-computation attenuation correction processing; correspondingly, the second PET data represents a set of PET images of a conventional area, and has lower computing power requirements.

[0065] Specifically, in some specific embodiments, in a processing scenario including chest and abdominal images, the converted CT data is used to determine the upper and lower boundaries of the chest and abdomen through manual identification and division, or through manual annotation and AI automatic identification and division; wherein, the upper and lower refer to continuous multi-layer CT images, from which layer the chest and abdominal tissue begins to appear, and from which layer the chest and abdominal tissue ends; because the PET image is blurry and it is not easy to distinguish the upper and lower boundaries of the chest and abdomen, it is necessary to generate virtual CT data for judgment, because the CT data is clearer.

[0066] S400, performing a first attenuation correction on the first PET data using the three-dimensional model to obtain first corrected data;

[0067] The three-dimensional model is generated by training the PET data of the paired three-dimensional blocks; it should be noted that in some embodiments, the step of obtaining the three-dimensional model by training the PET data of the paired three-dimensional blocks is also included, and this step includes: obtaining PET data; performing three-dimensional block extraction on the PET data by a three-dimensional extraction component, and converting to obtain PET data in a three-dimensional block format; constructing a first residual network; the first residual network includes a first feature extraction module, a first feature encoding module, a first feature capture module and a first normalization module; using the PET data in the three-dimensional block format as a training data set, and using the PET data of the paired three-dimensional blocks as label data; training and optimizing the first residual network through the training data set and the label data, and adjusting the first residual network based on the training results to obtain the three-dimensional model. In the process of training the three-dimensional model, in order to ensure the training effect, the applied PET data usually uses none-AC PET data (i.e., data without attenuation correction), and the first feature extraction module, the first feature encoding module and the first feature capture module all include a number of convolutional layers, and every two convolutional layers are connected using residual links.

[0068] In some embodiments, a three-dimensional model is used to perform a first attenuation correction on the first PET data to obtain first corrected data, including: performing a first feature extraction on the first PET data to obtain a first low-level feature; performing a first feature encoding based on the first low-level feature to obtain a first intermediate feature; performing a first feature capture based on the first intermediate feature to obtain a first high-level feature; and performing a first normalization process on the first high-level feature to obtain the first corrected data.

[0069] Specifically, in some specific embodiments, the none-AC PET data of the chest and abdomen are input into a 3D deep learning network model (i.e., a three-dimensional model), and the PET data of the chest and abdomen after attenuation correction (i.e., first corrected data) are output; wherein, the 3D deep learning network model needs to be trained in advance; since the chest and abdomen are highly heterogeneous regions, the use of a 3D deep learning network model can further improve the accuracy of attenuation correction compared to a 2D deep learning network model;

[0070] 3D deep learning network model G 3D Training process:

[0071] 1. Training data set: Input: chest and abdomen none-AC PET converted into 3D patches format by 3D Patch extraction, label: paired 3D patch AC PET;

[0072] 2. Perform model training optimization based on training data sets and labels, and make adjustments based on training results to obtain a 3D deep learning network model; model output: synthetic chest and abdominal AC PET;

[0073] The 3D deep learning network model can use the residual network (ResNet), such as Figure 4 As shown, the symbols in the figure are explained as follows: Conv: convolution, Kr: convolution kernel, BN: batch normalization, LReLU: Leaky ReLU activation function, Softmax: normalized exponential function, Dlt: dilation coefficient, Residual: residual block. The network model consists of 20 convolutional layers. The first 7 layers (constituting the first feature extraction module) use 3x3x3 convolution to extract low-level features (low-level features) such as edges of the input data; the following 6 layers (constituting the first feature encoding module) use dilated convolution with a dilation factor of 2 to encode the medium-level features (medium-level features) of the input data, and the next 6 layers (constituting the first feature capture module) use dilated convolution with a dilation factor of 4 to capture high-level features (high-level features), and the last layer (constituting the first normalization module) uses 1x1x1 convolution. Among them, the residual link connects every two convolutional layers. Batch Normalization and Leaky ReLU activation functions are connected to the convolutional layers located in the residual block. As shown Figure 5 As shown, through training, a 3D deep learning network model G is obtained from the none-AC PET data of the chest and abdomen to the attenuation-corrected PET data of the chest and abdomen. 3D , the scanned chest and abdomen none-AC PET data can be directly input into the trained model G in a three-dimensional block format 3D , and obtain the synthesized attenuation-corrected PET data of the chest and abdomen.

[0074] S500, performing a second attenuation correction on the second PET data using the two-dimensional model to obtain second corrected data;

[0075] Wherein, the two-dimensional model is generated by training the paired two-dimensional sequence of PET data; it should be noted that, in some embodiments, the step of obtaining the two-dimensional model by training the paired two-dimensional sequence of PET data is also included, and this step includes: obtaining PET data; wherein the PET data includes continuous multiple layers of PET images, and based on this characteristic, the PET data is usually in a two-dimensional sequence format; constructing a second residual network; the second residual network includes a second feature extraction module, a second feature encoding module, a second feature capture module and a second normalization module; using the PET data in the two-dimensional sequence format as a training data set, and using the paired two-dimensional sequence of PET data as label data; training and optimizing the second residual network through the training data set and the label data, and adjusting the second residual network based on the training results to obtain the two-dimensional model. Wherein, in the two-dimensional model training process, in order to ensure the training effect, the applied PET data usually uses none-AC PET data (i.e., data without attenuation correction), and the second feature extraction module, the second feature encoding module and the second feature capture module all include a number of convolutional layers, and every two convolutional layers are connected using residual links.

[0076] In some embodiments, a two-dimensional model is used to perform a second attenuation correction on the second PET data to obtain second corrected data, including: performing a second feature extraction on the second PET data to obtain a second low-level feature; performing a second feature encoding based on the second low-level feature to obtain a second intermediate feature; performing a second feature capture based on the second intermediate feature to obtain a second high-level feature; and performing a second normalization process on the second high-level feature to obtain second corrected data.

[0077] Specifically, in some specific embodiments, the none-AC PET data other than the chest and abdomen (i.e., the second PET data) is input into a 2D deep learning network model (i.e., a two-dimensional model), and the PET data other than the chest and abdomen after attenuation correction (i.e., the second corrected data) is output;

[0078] The 2D deep learning network model needs to be trained in advance; the 2D deep learning network model is sufficient to meet the attenuation correction accuracy, and there is no need to use a 3D deep learning network model; the 2D deep learning network model G 2D Training process:

[0079] 1. Training data set: Input: none-AC PET other than chest and abdomen in 2D slices format, label: paired 2D sequence AC PET;

[0080] 2. Perform model training optimization based on training data sets and labels, and make adjustments based on training results to obtain a 2D deep learning network; model output: synthetic AC PET other than chest and abdomen;

[0081] 2D deep learning network models can also use residual networks (ResNet), such as Figure 4 As shown in the figure, the network model consists of 20 convolutional layers. The first 7 layers (constituting the second feature extraction module) use 3x3x3 convolution to extract low-level features of the input data such as edges; the next 6 layers (constituting the second feature encoding module) use dilated convolution with a dilation factor of 2 to encode the medium-level features of the input data, and the next 6 layers (constituting the second feature capture module) use dilated convolution with a dilation factor of 4 to capture high-level features. The last layer (constituting the first normalization module) uses 1x1x1 convolution. Among them, the residual link connects every two convolutional layers. Batch Normalization and Leaky ReLU activation functions are connected to the convolutional layers located in the residual block. As shown Figure 6 As shown, through training, a 2D deep learning network model G is obtained from the none-AC PET data outside the chest and abdomen to the attenuation-corrected PET data outside the chest and abdomen. 2D , the scanned none-AC PET data other than the chest and abdomen can be directly input into the trained model G in a two-dimensional sequence format 2D , and obtain the synthesized attenuation-corrected PET data other than the chest and abdomen.

[0082] S600, obtaining target PET data according to the first correction data and the second correction data;

[0083] It should be noted that, in some embodiments, based on the boundary corresponding to the grouping result, the first corrected data and the second corrected data are spliced ​​and fused to obtain the target PET data.

[0084] Specifically, in some specific embodiments, the aforementioned attenuation-corrected PET data of the chest and abdomen and the attenuation-corrected PET data other than the chest and abdomen are spliced ​​and fused to form attenuation-corrected whole-body PET data.

[0085] In summary, the embodiments of the present invention can more conveniently and quickly implement PET attenuation correction through module-based deep learning network model training, thereby realizing direct attenuation correction technology (direct-AC) of PET images based on deep learning. It does not require additional scanning of anatomical images (CT or MRI), but directly generates attenuation-corrected PET images (AC PET) directly from PET images without attenuation correction (none-AC PET) through deep learning algorithms. Compared with traditional CTAC, direct-AC technology can reduce the radiation dose of additional CT scans, and is suitable for single PET scanners without CT or MRI, so it has good research and application prospects. The error of the Direct-AC method mainly depends on the accuracy of the deep neural network itself. On the one hand, a large amount of data is required for deep learning network model training. On the other hand, current DLAC research mainly focuses on large bone structures with relatively simple shapes such as the brain and pelvis. For example, it is difficult to balance the accuracy of attenuation correction of the chest and other parts of the body. This is also a major challenge faced by current PET direct attenuation correction based on deep learning. Based on this, the embodiments of the present invention further distinguish the chest and abdomen and use different network learning models to achieve the optimal attenuation correction effect throughout the body. The embodiments of the present invention provide a new method for direct attenuation correction of whole-body PET images based on deep learning with better generalization ability and higher accuracy, which is in line with the development trend of PET clinical applications and has great research value.

[0086] On the other hand, an embodiment of the present invention provides a PET attenuation correction system, comprising: a first module, used to acquire PET data to be corrected; the PET data to be corrected includes continuous multi-layer PET images; a second module, used to use a diffusion model to perform data conversion on the PET data to be corrected to obtain CT data; wherein the diffusion model is generated by training PET data of paired CT data; the CT data includes continuous multi-layer CT images; a third module, used to determine the grouping result of the CT data based on the marked boundary, and divide the PET data to be corrected into first PET data and second PET data according to the grouping result; a fourth module, used to use a three-dimensional model to perform a first attenuation correction on the first PET data to obtain first corrected data; wherein the three-dimensional model is generated by training PET data of paired three-dimensional blocks; a fifth module, used to use a two-dimensional model to perform a second attenuation correction on the second PET data to obtain second corrected data; wherein the two-dimensional model is generated by training PET data of paired two-dimensional sequences; a sixth module, used to obtain target PET data based on the first corrected data and the second corrected data.

[0087] The contents of the method embodiments of the present invention are all applicable to the system embodiments. The functions specifically implemented by the system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method.

[0088] Another aspect of an embodiment of the present invention further provides a PET attenuation correction device, including a processor and a memory;

[0089] The memory is used to store programs;

[0090] The processor executes the program to implement the above method.

[0091] The contents of the method embodiments of the present invention are all applicable to the device embodiments. The functions specifically implemented by the device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0092] Another aspect of the embodiments of the present invention further provides a computer-readable storage medium, wherein the storage medium stores a program, and the program is executed by a processor to implement the above method.

[0093] The contents of the method embodiments of the present invention are all applicable to the computer-readable storage medium embodiments. The functions specifically implemented by the computer-readable storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0094] The embodiment of the present invention also discloses a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device can read the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the above method.

[0095] In some selectable embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the present invention is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Selectable embodiments are expected, wherein the order of various operations is changed and the sub-operation of a part for which is described as a larger operation is performed independently.

[0096] In addition, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise specified, one or more of the functions and / or features can be integrated into a single physical device and / or software module, or one or more functions and / or features can be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present invention. More specifically, in view of the properties, functions and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional skills of the engineer. Therefore, those skilled in the art can implement the present invention set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.

[0097] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0098] The logic and / or steps represented in the flowchart or otherwise described herein, for example, may be considered as an ordered list of executable instructions for implementing logical functions, and may be embodied in any computer-readable medium for use by an instruction execution device, apparatus or device (such as a computer-based device, a device including a processor, or other device that can fetch instructions from an instruction execution device, apparatus or device and execute the instructions), or in conjunction with such instruction execution device, apparatus or device. For purposes of this specification, "computer-readable medium" may be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution device, apparatus or device, or in conjunction with such instruction execution device, apparatus or device.

[0099] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0100] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution device. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0101] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0102] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

[0103] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. A PET attenuation correction method, characterized in that: include: Acquire PET data to be corrected; the PET data to be corrected includes continuous multi-layer PET images; Using a diffusion model, the PET data to be corrected is converted to obtain CT data; wherein the diffusion model is generated by training PET data paired with CT data; the CT data includes continuous multi-layer CT images; Determine a grouping result of the CT data based on the marked boundaries, and divide the PET data to be corrected into first PET data and second PET data according to the grouping result; the marked boundaries represent the uppermost and lowermost boundaries of the chest and abdomen, the first PET data represent a PET image set of a highly heterogeneous area, and the second PET data represent a PET image set of a conventional area; Using a three-dimensional model, performing a first attenuation correction on the first PET data to obtain first corrected data; wherein the three-dimensional model is generated by training the PET data of paired three-dimensional blocks; and the three-dimensional model represents a 3D deep learning network model; Using the two-dimensional model, performing a second attenuation correction on the second PET data to obtain second corrected data; wherein the two-dimensional model is generated by training the PET data of the paired two-dimensional sequence; and the two-dimensional model represents a 2D deep learning network model; Target PET data is obtained according to the first correction data and the second correction data.

2. A PET attenuation correction method according to claim 1, characterized in that: The method of using the diffusion model to convert the PET data to be corrected to obtain CT data includes: Using a diffusion model, forward diffusion processing is performed on each of the PET images of the PET data to be corrected to obtain Gaussian noise; According to the Gaussian noise, a CT image corresponding to the PET image is restored through reverse sampling processing, and the CT image is integrated to obtain CT data.

3. A PET attenuation correction method according to claim 1, characterized in that: The step of obtaining a three-dimensional model by training the PET data of the paired three-dimensional blocks includes: Acquire PET data; Performing three-dimensional block extraction on the PET data by a three-dimensional extraction component, and converting to obtain PET data in a three-dimensional block format; Constructing a first residual network; the first residual network includes a first feature extraction module, a first feature encoding module, a first feature capture module and a first normalization module; Using the PET data in the three-dimensional block format as a training data set, and using the PET data of the paired three-dimensional blocks as label data; The first residual network is trained and optimized using the training data set and the label data, and based on the training result, the first residual network is adjusted to obtain a three-dimensional model.

4. A PET attenuation correction method according to claim 1, characterized in that: The step of performing a first attenuation correction on the first PET data using the three-dimensional model to obtain first corrected data includes: The first PET data is subjected to first feature extraction, first feature encoding, first feature capture and first normalization processing to obtain first corrected data.

5. A PET attenuation correction method according to claim 1, characterized in that: The step of obtaining a two-dimensional model by training the PET data of the paired two-dimensional sequence includes: Acquire PET data; wherein the PET data includes continuous multi-layer PET images, and the PET data is in a two-dimensional sequence format; Constructing a second residual network; the second residual network includes a second feature extraction module, a second feature encoding module, a second feature capture module and a second normalization module; Using the PET data in the two-dimensional sequence format as a training data set, and using the PET data in the paired two-dimensional sequence as label data; The second residual network is trained and optimized using the training data set and the label data, and based on the training result, the second residual network is adjusted to obtain a two-dimensional model.

6. A PET attenuation correction method according to claim 1, characterized in that: The step of performing a second attenuation correction on the second PET data using the two-dimensional model to obtain second corrected data includes: The second PET data is subjected to second feature extraction, second feature encoding, second feature capture and second normalization processing to obtain second corrected data.

7. A PET attenuation correction method according to claim 1, characterized in that: The step of obtaining target PET data according to the first correction data and the second correction data includes: Based on the boundary corresponding to the grouping result, the first corrected data and the second corrected data are spliced ​​and fused to obtain target PET data.

8. A PET attenuation correction system, characterized in that: include: The first module is used to obtain PET data to be corrected; the PET data to be corrected includes continuous multi-layer PET images; The second module is used to perform data conversion on the PET data to be corrected by using a diffusion model to obtain CT data; wherein the diffusion model is generated by training PET data paired with CT data; and the CT data includes continuous multi-layer CT images; A third module is used to determine a grouping result of the CT data based on the marked boundaries, and divide the PET data to be corrected into first PET data and second PET data according to the grouping result; the marked boundaries represent the uppermost and lowermost boundaries of the chest and abdomen, the first PET data represent a PET image set of a highly heterogeneous area, and the second PET data represent a PET image set of a conventional area; A fourth module is used to perform a first attenuation correction on the first PET data using a three-dimensional model to obtain first corrected data; wherein the three-dimensional model is generated by training the PET data of paired three-dimensional blocks; and the three-dimensional model represents a 3D deep learning network model; A fifth module is used to perform a second attenuation correction on the second PET data using a two-dimensional model to obtain second corrected data; wherein the two-dimensional model is generated by training the PET data of a paired two-dimensional sequence; and the two-dimensional model represents a 2D deep learning network model; The sixth module is used to obtain target PET data according to the first correction data and the second correction data.

9. A PET attenuation correction device, comprising a processor and a memory; The memory is used to store programs; The processor executes the program to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The storage medium stores a program, and the program is executed by a processor to implement the method according to any one of claims 1 to 7.

Images