PET (positron emission tomography) image generation method based on ultrahigh-field MRI (magnetic resonance imaging) and related equipment

Abstract

The invention discloses a PET (positron emission tomography) image generation method based on ultrahigh-field MRI (magnetic resonance imaging) and related equipment. The method comprises the following steps: acquiring a target image, wherein the target image is an ultrahigh-field MRI image; a target joint score is obtained, the initial image is processed multiple times based on the target joint score and the target image to obtain a target PET image, and each processing comprises sampling processing and correction processing, each processing process comprises the steps of sampling the initial image based on the target image in a backward process of a diffusion model according to the target joint score, and then correcting the initial image based on the target image according to an annealing Rangevin equation. According to the PET image generation method based on the ultrahigh-field MRI provided by the invention, the method for generating the corresponding PET image by using the ultrahigh-field MRI image is realized, so that the abnormality of the tissue of the scanned part of the patient can be detected earlier.

Description

Technical Field

[0001] The present invention relates to the technical field of text detection, and in particular to a PET image generation method based on ultra-high field MRI and related equipment. Background Art

[0002] Multimodal medical images in the prior art, such as magnetic resonance imaging and positron emission tomography, can provide a variety of information and have been widely used in the diagnosis of neurodegenerative diseases. Among them, MRI images are mainly used to detect structural changes in organs, while PET images are mainly used to detect functional changes in organs. PET images can detect tissue abnormalities earlier than MRI images. However, due to limitations such as cost, radiation, and magnetic fields, the acquired PET images may be missing. In order to solve this problem, image generation technology is usually used to generate PET images using MRI images.

[0003] Currently, the image generation task can generate corresponding PET images using high-field MRI images (3T MRI). Since higher-resolution MRI images can detect tissue abnormalities more quickly, they need to be generated using ultra-high-field MRI (5T MRI or 7T MRI) images. However, in the existing technology, it is not possible to generate corresponding PET images using ultra-high-field MRI images.

[0004] Therefore, the prior art still needs to be improved and enhanced. Summary of the invention

[0005] In view of the above-mentioned defects of the prior art, a PET image generation method based on ultra-high field MRI and related equipment are provided, aiming to solve the problem that the prior art cannot generate corresponding PET images according to ultra-high field MRI images.

[0006] A first aspect of the present invention provides a method for generating a PET image based on ultra-high field MRI, comprising:

[0007] Acquiring a target image, wherein the target image is an ultra-high field MRI image;

[0008] Obtaining a target joint score, where the target joint score is a derivative of a logarithm of a joint probability distribution of a first image and a second image in a forward process of a diffusion model, where the first image and the second image are medical images taken for the same receptor;

[0009] The initial image is processed multiple times based on the target joint score and the target image to obtain a target PET image, wherein the initial image is a pure Gaussian noise image, each processing includes sampling processing and correction processing, and each processing process includes:

[0010] The initial image is sampled based on the target image in a backward process of a diffusion model according to the target joint score, and then the initial image is corrected based on the target image according to an annealed Langevin equation.

[0011] The method for generating PET images based on ultra-high field MRI, wherein the step of obtaining a target joint score comprises:

[0012] Acquire the first image and the second image, wherein the first image is an MRI image and the second image is a PET image;

[0013] The second image is processed multiple times based on the first image until the second image becomes a pure Gaussian noise image, thereby obtaining multiple target joint scores.

[0014] The method for generating a PET image based on ultra-high field MRI, wherein the step of processing the second image multiple times based on the first image comprises:

[0015] adding Gaussian noise to the second image;

[0016] Inputting the first image and the second image after adding Gaussian noise into a target neural network for deep learning;

[0017] Repeat the steps of adding Gaussian noise to the second image and deep learning to obtain a target joint score.

[0018] The method for generating PET images based on ultra-high field MRI, wherein the step of adding Gaussian noise to the second image comprises:

[0019] adding Gaussian noise to the second image according to a first formula;

[0020] The first formula is:

[0021] x i =x i-1 +g i-1 z;

[0022] Where i = 1, 2, ..., N, N is a positive integer, g i-1 is the noise coefficient of the i-1th time Gaussian noise is added, z is Gaussian noise, x i is the second image after adding Gaussian noise for the i-1th time, x i-1 It is the second image after Gaussian noise is added for the i-2th time.

[0023] The method for generating PET images based on ultra-high field MRI, wherein the first image and the second image after adding Gaussian noise are input into a target neural network for deep learning to obtain a target joint score, comprises:

[0024] The target joint score is obtained by solving the derivative of the logarithm of the joint probability distribution for the first image and the second image after Gaussian noise is added.

[0025] The method for generating PET images based on ultra-high field MRI, wherein the initial image is sampled based on the target image in a backward process of a diffusion model according to the target joint score, comprises:

[0026] Sampling the initial image according to a second formula;

[0027] The second formula is:

[0028]

[0029] Where j = 1, 2, ..., N + 1, is the function corresponding to the target joint score obtained after the second image is processed for the N-j+1th time based on the first image, g j is the noise coefficient of adding Gaussian noise to the second image for the N-j+1th time, x′ j+1 is the initial image after the jth sampling, t′ is the target image, f j (x′ j ) is the drift coefficient.

[0030] The method for generating a PET image based on ultra-high field MRI, wherein the correction of the initial image based on the target image according to the annealed Langevin equation comprises:

[0031] Correcting the initial image according to a third formula;

[0032] The third formula is:

[0033]

[0034] where ε=2(r||z|| 2 / ||s θ || 2 ), r is the parameter, x" j+1 is the initial image after the j-th correction.

[0035] A second aspect of the present invention provides a PET image generation device based on ultra-high field MRI, comprising:

[0036] An image acquisition module, wherein the image acquisition module is used to acquire a target image, wherein the target image is an ultra-high field MRI image;

[0037] A joint score acquisition module, the joint score acquisition module is used to acquire a target joint score, the target joint score is a derivative of a logarithm of a joint probability distribution of a first image and a second image in a forward process of a diffusion model, the first image and the second image being medical images taken for the same receptor;

[0038] The image generation module processes the initial image multiple times based on the target joint score and the target image to obtain a target PET image, wherein the initial image is a pure Gaussian noise image, each processing includes sampling processing and correction processing, and each processing process includes:

[0039] The initial image is sampled based on the target image in a backward process of a diffusion model according to the target joint score, and then the initial image is corrected based on the target image according to an annealed Langevin equation.

[0040] According to a third aspect of the present invention, a terminal is provided, comprising a processor and a computer-readable storage medium communicatively connected to the processor, wherein the computer-readable storage medium is suitable for storing a plurality of instructions, and the processor is suitable for calling the instructions in the computer-readable storage medium to execute the steps of implementing any one of the above-mentioned methods for generating PET images based on ultra-high field MRI.

[0041] A fourth aspect of the present invention provides a computer-readable storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the PET image generation method based on ultra-high field MRI as described in any of the above.

[0042] Beneficial effects: Compared with the prior art, the present invention provides a PET image generation method based on ultra-high field MRI and related equipment. In the PET image generation method based on ultra-high field MRI provided by the present invention, a target image is acquired, which is an ultra-high field MRI image, and then a target joint score is acquired. The target joint score is the derivative of the logarithm of the joint probability distribution of a first image and a second image in the forward process of a diffusion model, and the first image and the second image are medical images taken for the same receptor. Based on the target joint score and the target image, the initial image is processed multiple times to obtain a target PET image, wherein the initial image is a pure Gaussian noise image, and each processing includes sampling processing and correction processing. Each processing process includes: sampling the initial image based on the target image in the backward process of the diffusion model according to the target joint score, and then correcting the initial image based on the target image according to the annealed Langevin equation. The PET image generation method based on ultra-high field MRI provided by the present invention only needs to learn according to multiple groups of the first images and the second images in the forward process of the diffusion model in the early stage to obtain the target joint score, and then the ultra-high field MRI image can be combined with a pure Gaussian noise image based on the target joint score to obtain a high-definition PET image, thereby realizing a method for generating corresponding PET images using ultra-high field MRI images, so that abnormalities in the tissue of the scanned part of the patient can be detected earlier. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A flowchart of an embodiment of a method for generating a PET image based on ultra-high field MRI provided by the present invention;

[0044] Figure 2 A prior art model diagram of an embodiment of a PET image generation method based on ultra-high field MRI provided by the present invention;

[0045] Figure 3 A noise processing diagram in an embodiment of a PET image generation method based on ultra-high field MRI provided by the present invention;

[0046] Figure 4 An example diagram of the processing process in an embodiment of the method for generating PET images based on ultra-high field MRI provided by the present invention;

[0047] Figure 5 A result comparison diagram of an embodiment of the PET image generation method based on ultra-high field MRI provided by the present invention;

[0048] Figure 6 Example of results in an embodiment of the method for generating PET images based on ultra-high field MRI provided by the present invention Figure 1 ;

[0049] Figure 7 Example of results in an embodiment of the method for generating PET images based on ultra-high field MRI provided by the present invention Figure 2 ;

[0050] Figure 8 A schematic structural diagram of an embodiment of a PET image generation device based on ultra-high field MRI provided by the present invention;

[0051] Fig. 9 A structural principle diagram of an embodiment of a terminal provided by the present invention. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0053] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "one", "said", and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0054] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as herein.

[0055] The present invention provides a PET image generation method based on ultra-high field MRI, which can be applied to a terminal with computing capability. The terminal can execute the PET image generation method based on ultra-high field MRI provided by the present invention to operate the user space in the target system.

[0056] Embodiment 1

[0057] In this embodiment, a method for generating PET images based on ultra-high field MRI is provided. Figure 1 As shown, the PET image generation method based on ultra-high field MRI provided by the present invention comprises the steps of:

[0058] S100: Acquire a target image, where the target image is an ultra-high field MRI image.

[0059] Specifically, the target image in this embodiment is an ultra-high field MRI image. In the prior art, there are already methods for generating corresponding PET images using high-field MRI images (3T MRI), but generating corresponding PET images using ultra-high field MRI images has not yet been achieved. Although, due to the emergence of adversarial generative networks, the possibility of generating corresponding PET images using MRI images has increased. Specifically, compared with traditional variational generative models, higher quality PET images can be generated based on adversarial networks. As a method for generating high-quality images, the adversarial network can be described as two parts, including a generation network and a discrimination network. First, the PET image generated by the generation network is then authenticated by the discrimination network to determine whether it is true or not, and then the authentication result is fed back to the generation network for optimization. However, the PET generated by the generation network cannot be identified as true or false by the discrimination network, such as Figure 2 That is to say, the generative model in the prior art is prone to mode collapse during the training process due to adversarial training, and the image generation lacks diversity, and it is also impossible to generate corresponding PET images from ultra-high field MRI images. However, the PET image generation method based on ultra-high field MRI provided in this embodiment can generate corresponding PET images based on the ultra-high field MRI image of the head as a guide through the diffusion model of the joint probability distribution.

[0060] Specifically, after acquiring the corresponding ultra-high field MRI image, it needs to be processed, and the steps are as follows:

[0061] S200, obtaining a target joint score, where the target joint score is a derivative of a logarithm of a joint probability distribution of a first image and a second image in a forward process of a diffusion model, where the first image and the second image are medical images taken for the same receptor.

[0062] Wherein, obtaining the target joint score includes:

[0063] S210: Acquire the first image and the second image, wherein the first image is an MRI image and the second image is a PET image.

[0064] Specifically, firstly, a set of registration data is obtained for extracting the target joint parameters. The set of registration data is MRI and PET images of the same receptor. In this embodiment, the MRI image of the same receptor is defined as the first image, and the PET image of the same receptor is defined as the second image.

[0065] S220. Process the second image multiple times based on the first image until the second image becomes a pure Gaussian noise image, and obtain multiple target joint scores.

[0066] The performing multiple processing on the second image based on the first image includes:

[0067] S221. Add Gaussian noise to the second image.

[0068] The adding Gaussian noise to the second image comprises:

[0069] adding Gaussian noise to the second image according to a first formula;

[0070] The first formula is:

[0071] x i =x i-1 +g i-1 z, i = 1, 2, ..., N;

[0072] Among them, g i-1 is the noise coefficient of the i-1th time Gaussian noise is added, N is a positive integer, z is Gaussian noise, x i is the second image after adding Gaussian noise for the i-1th time, x i-1 It is the second image after Gaussian noise is added for the i-2th time.

[0073] Specifically, That is to say, the first formula can also be expressed as:

[0074]

[0075] Among them, σ 1 , σ 2 , …, σ i-1 , σ i , …, σ N is a parameter value determined according to actual needs, z is Gaussian noise, x i is the second image after adding Gaussian noise for the i-1th time, x i-1 It is the second image after Gaussian noise is added for the i-2th time.

[0076] S222, inputting the first image and the second image after adding Gaussian noise into a target neural network for deep learning;

[0077] S223. Repeat the steps of adding Gaussian noise and deep learning to the second image to obtain a target joint score.

[0078] The step of inputting the first image and the second image after adding Gaussian noise into a target neural network for deep learning to obtain a target joint score includes:

[0079] The target joint score is obtained by solving the derivative of the logarithm of the joint probability distribution for the first image and the second image after Gaussian noise is added.

[0080] Specifically, when obtaining the target joint score, it is necessary to first prepare an MRI image and a PET image of the same receptor, and determine that the MRI image and the PET image are registration data. In this embodiment, the MRI image is defined as the first image, and the PET image is defined as the second image. Repeat the steps of adding Gaussian noise to the second image and deep learning to obtain the target joint score.

[0081] Reference Figure 3 , add Gaussian noise to the second image multiple times until the second image becomes Gaussian noise. The process of adding noise to the second image is:

[0082]

[0083] Where i = 1, 2, ..., N, N is a positive integer, σ i is a parameter value determined according to actual needs, z is Gaussian noise, x i is the second image after adding Gaussian noise for the i-1th time, x i-1 This is the second image after Gaussian noise is added for the (i-2)th time. In this embodiment, N is 1000, which is equivalent to adding Gaussian noise for N-1 times, that is, 999 times in total.

[0084] After each noise addition, the target neural network needs to learn the derivative of the logarithm of the joint probability distribution of the first image (MRI image) and the second image (PET image) after the noise addition, that is, the joint score, where the joint score obtained after the i-1th noise addition can be expressed as a function Among them, x iis the second image after adding Gaussian noise for the i-1th time, y is the first image, and i=1, 2, ..., N. In this embodiment, the target network is a deep neural network (U-Net) for encoding and decoding, that is, in this embodiment, the second image needs to be subjected to 999 noise addition processes, and each noise addition process is followed by learning through the target neural network to obtain a joint score, and then the joint score obtained by adding the original second image and the first image is added, so there are a total of 1000 joint scores, and then based on the 1000 joint scores, the target image is sampled in the inverse diffusion model.

[0085] S300, performing multiple processing on the initial image based on the target joint score and the target image to obtain a target PET image, wherein the initial image is a pure Gaussian noise image, each processing includes sampling processing and correction processing, and each processing process includes:

[0086] The initial image is sampled based on the target image in a backward process of a diffusion model according to the target joint score, and then the initial image is corrected based on the target image according to an annealed Langevin equation.

[0087] Specifically, the 1000th joint score is combined with the target image and the initial image, that is, a pure Gaussian noise image, for sampling and correction to obtain the initial image after the first processing, and the 999th joint score is combined with the target image and the initial image after the first processing to sample and correct, to obtain the initial image after the second processing, and so on, a total of 1000 sampling and correction are performed on the initial image, and the final image obtained is the target PET image.

[0088] Reference Figure 4 After obtaining the joint score, the target image is sampled multiple times in the backward process of the diffusion model according to the target joint score, and the target image is corrected according to the annealed Langevin equation after each sampling.

[0089] Specifically, sampling the initial image based on the target image in a backward process of the diffusion model according to the target joint score includes:

[0090] S310, sampling the initial image according to a second formula;

[0091] The second formula is:

[0092]

[0093] Where j = 1, 2, ..., N + 1, is the function corresponding to the target joint score obtained after the second image is processed for the N-j+1th time based on the first image, g j is the noise coefficient of adding Gaussian noise to the second image for the N-j+1th time, x′ j+1 is the initial image after the jth sampling, y′ is the target image, and f j (x′ j ) is the drift coefficient.

[0094] Specifically, in the backward process stage of the diffusion model, data is sampled based on the target joint score using the second formula.

[0095] The second formula is:

[0096]

[0097] Where j = 1, 2, ..., N + 1, is the function corresponding to the target joint score obtained by processing the second image for the N-j+1th time based on the first image, g j is the noise coefficient of adding Gaussian noise to the second image for the N-j+1th time, x′ j+1 is the initial image after the jth sampling, y′ is the target image, and f j (x′ j ) is the drift coefficient. Specifically, Among them, σ 1 , σ 2 , …, σ N-j+1 , σ N-j , …, σ N It is a parameter value determined according to actual needs.

[0098] It is not difficult to see that when the initial image is processed for the first time based on the target joint score and the target image, the function corresponding to the target joint score obtained after the second image is processed for the N-1th time based on the first image is substituted into the second formula, and when the initial image is processed for the second time based on the target joint score and the target image, the function corresponding to the target joint score obtained after the second image is processed for the N-2th time based on the first image is substituted into the second formula, and so on, until all the target joint scores are used up.

[0099] In this embodiment, after each sampling is completed, the initial image needs to be corrected based on the target image according to the annealed Langevin equation.

[0100] S320, correcting the initial image based on the target image according to the annealed Langevin equation, comprising:

[0101] Correcting the initial image according to a third formula;

[0102] The third formula is:

[0103]

[0104] in, is the parameter, x" j+1 is the initial image after the j-th correction.

[0105] When the last correction is completed, the initial image obtained is the target PET image. The target PET image is generated by the PET image generation method based on ultra-high field MRI provided by this embodiment and based on the backward process of the diffusion model.

[0106] That is to say, the method provided in this embodiment only needs to learn according to multiple groups of the first image and the second image in the forward process of the diffusion model in the early stage, obtain the target joint score, and then save the target joint score, so that it can be called in the backward process of the diffusion model during the later conversion, without the need to re-learn. For example, in this embodiment, N is 1000, then there are 1000 target joint scores saved in the diffusion model, and when it is necessary to convert the target image, that is, any high-field MRI image, to obtain a PET image, these 1000 target joint scores can be called. Specifically, the target joint score is a function that can be brought into the MRI image to be converted, and different results can be obtained according to different MRI images.

[0107] like Figure 5 As shown, the PET image generation method based on ultra-high field MRI provided in this embodiment takes into account the diversity and stability of the generation model. While reducing the computational complexity, the positron tomography scan generated by magnetic resonance guidance is more diverse than the positron tomography scan generated by other methods, and the robustness of the model is enhanced.

[0108] Through simulation and experiment, it is proved that the PET image generation method based on ultra-high field MRI provided in this embodiment has a higher signal-to-noise ratio for the positron tomography scan generated by ultra-high field nuclear magnetic resonance (MRI) than for the positron tomography scan (PET) generated by high field nuclear magnetic resonance (MRI), and at the same time, the positron tomography scan generated by ultra-high long nuclear magnetic resonance has a higher signal-to-noise ratio than the generation method using the adversarial network. Figure 6 , Figure 7 shown.

[0109] In summary, this embodiment provides a PET image generation method based on ultra-high field MRI, by acquiring a target image, which is an ultra-high field MRI image, and then acquiring a target joint score, wherein the target joint score is the derivative of the logarithm of the joint probability distribution of the first image and the second image in the forward process of the diffusion model, and the first image and the second image are medical images taken for the same receptor. The initial image is processed multiple times based on the target joint score and the target image to obtain a target PET image, wherein the initial image is a pure Gaussian noise image, and each processing includes sampling processing and correction processing, and each processing process includes: sampling the initial image based on the target image in the backward process of the diffusion model according to the target joint score, and then correcting the initial image based on the target image according to the annealed Langevin equation. The method for generating PET images based on ultra-high field MRI provided in this embodiment only requires, in the early stage, to learn based on multiple groups of the first images and the second images in the forward process of the diffusion model to obtain the target joint score, and then the ultra-high field MRI image can be combined with a pure Gaussian noise image based on the target joint score to obtain a high-definition PET image, thereby realizing a method for generating corresponding PET images using ultra-high field MRI images, so that abnormalities in the tissue of the scanned part of the patient can be detected earlier.

[0110] It should be understood that, although the various steps in the flowcharts given in the accompanying drawings of the present invention are displayed in sequence according to the indications of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless there is a clear description in this article, the execution of the steps in the present invention is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps of the present invention may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0111] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0112] Embodiment 2

[0113] Based on the above embodiments, the present invention further provides a PET image generation device based on ultra-high field MRI, the functional module diagram of which is shown in FIG. Figure 8 As shown, the PET image generation device based on ultra-high field MRI includes:

[0114] An image acquisition module, the image acquisition module is used to acquire a target image, the target image is an ultra-high field MRI image, as specifically described in the first embodiment;

[0115] A joint score acquisition module, the joint score acquisition module is used to acquire a target joint score, the target joint score is a derivative of a logarithm of a joint probability distribution of a first image and a second image in a forward process of a diffusion model, the first image and the second image being medical images taken for the same receptor;

[0116] The image generation module processes the initial image multiple times based on the target joint score and the target image to obtain a target PET image, wherein the initial image is a pure Gaussian noise image, each processing includes sampling processing and correction processing, and each processing process includes:

[0117] The initial image is sampled based on the target image in the backward process of the diffusion model according to the target joint score, and then the initial image is corrected based on the target image according to the annealed Langevin equation, as specifically described in the first embodiment.

[0118] Embodiment 3

[0119] Based on the above embodiments, the present invention also provides a terminal, such as Fig. 9 As shown, the terminal includes a processor 10 and a memory 20. Fig. 9 Only some components of the terminal are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0120] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory of the terminal. In other embodiments, the memory 20 may also be an external storage device of the terminal, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the terminal. Further, the memory 20 may also include both an internal storage unit of the terminal and an external storage device. The memory 20 is used to store application software and various types of data installed on the terminal. The memory 20 may also be used to temporarily store data that has been output or is to be output. In one embodiment, a PET image generation program 30 based on ultra-high field MRI is stored on the memory 20, and the PET image generation program 30 based on ultra-high field MRI can be executed by the processor 10, thereby realizing the PET image generation method based on ultra-high field MRI in the present application.

[0121] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor or other chip, used to run the program code or process data stored in the memory 20, such as executing the PET image generation method based on ultra-high field MRI.

[0122] In one embodiment, when the processor 10 executes the ultra-high field MRI-based PET image generation program 30 in the memory 20, the following steps are implemented:

[0123] Acquiring a target image, wherein the target image is an ultra-high field MRI image;

[0124] Obtaining a target joint score, where the target joint score is a derivative of a logarithm of a joint probability distribution of a first image and a second image in a forward process of a diffusion model, where the first image and the second image are medical images taken for the same receptor;

[0125] The initial image is processed multiple times based on the target joint score and the target image to obtain a target PET image, wherein the initial image is a pure Gaussian noise image, each processing includes sampling processing and correction processing, and each processing process includes:

[0126] The initial image is sampled based on the target image in a backward process of a diffusion model according to the target joint score, and then the initial image is corrected based on the target image according to an annealed Langevin equation.

[0127] Wherein, obtaining the target joint score includes:

[0128] Acquire the first image and the second image, wherein the first image is an MRI image and the second image is a PET image;

[0129] The second image is processed multiple times based on the first image until the second image becomes a pure Gaussian noise image, thereby obtaining multiple target joint scores.

[0130] The performing multiple processing on the second image based on the first image includes:

[0131] adding Gaussian noise to the second image;

[0132] Inputting the first image and the second image after adding Gaussian noise into a target neural network for deep learning;

[0133] Repeating the steps of adding Gaussian noise to the second image and deep learning to obtain a target joint score. The step of adding Gaussian noise to the second image includes:

[0134] adding Gaussian noise to the second image according to a first formula;

[0135] The first formula is:

[0136] x i =x i-1 +g i-1 z;

[0137] Where i = 1, 2, ..., N, N is a positive integer, g i-1 is the noise coefficient of the i-1th time Gaussian noise is added, z is Gaussian noise, x i is the second image after adding Gaussian noise for the i-1th time, xi-1 It is the second image after Gaussian noise is added for the i-2th time.

[0138] The step of inputting the first image and the second image after adding Gaussian noise into a target neural network for deep learning to obtain a target joint score includes:

[0139] The target joint score is obtained by solving the derivative of the logarithm of the joint probability distribution for the first image and the second image after Gaussian noise is added.

[0140] The step of sampling the initial image based on the target image in a backward process of a diffusion model according to the target joint score includes:

[0141] Sampling the initial image according to a second formula;

[0142] The second formula is:

[0143]

[0144] Where j = 1, 2, ..., N + 1, is the function corresponding to the target joint score obtained after the second image is processed for the N-j+1th time based on the first image, g j is the noise coefficient of adding Gaussian noise to the second image for the N-j+1th time, x′ j+1 is the initial image after the jth sampling, y′ is the target image, and f j (x′ j ) is the drift coefficient.

[0145] The correcting of the initial image based on the target image according to the annealed Langevin equation includes:

[0146] Correcting the initial image according to a third formula;

[0147] The third formula is:

[0148]

[0149] where ε=2(r||z|| 2 / ||s θ || 2 ), r is the parameter, x" j+1 is the initial image after the j-th correction.

[0150] Embodiment 4

[0151] The present invention also provides a storage medium storing one or more programs, which can be executed by one or more processors to implement the steps of the PET image generation method based on ultra-high field MRI described in the above embodiment.

[0152] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating PET images based on ultra-high field MRI, It is characterized in that include: Acquiring a target image, wherein the target image is an ultra-high field MRI image; Obtaining a target joint score, where the target joint score is a derivative of a logarithm of a joint probability distribution of a first image and a second image in a forward process of a diffusion model, where the first image and the second image are medical images taken for the same receptor; The initial image is processed multiple times based on the target joint score and the target image to obtain a target PET image, wherein the initial image is a pure Gaussian noise image, each processing includes sampling processing and correction processing, and each processing process includes: The initial image is sampled based on the target image in a backward process of a diffusion model according to the target joint score, and then the initial image is corrected based on the target image according to an annealed Langevin equation.

2. The method for generating PET images based on ultra-high field MRI according to claim 1, It is characterized in that The obtaining of the target joint score comprises: Acquire the first image and the second image, wherein the first image is an MRI image and the second image is a PET image; The second image is processed multiple times based on the first image until the second image becomes a pure Gaussian noise image, thereby obtaining multiple target joint scores.

3. The method for generating PET images based on ultra-high field MRI according to claim 2, It is characterized in that The performing multiple processing on the second image based on the first image comprises: adding Gaussian noise to the second image; Inputting the first image and the second image after adding Gaussian noise into a target neural network for deep learning; Repeat the steps of adding Gaussian noise to the second image and deep learning to obtain a target joint score.

4. The method for generating PET images based on ultra-high field MRI according to claim 3, It is characterized in that The adding Gaussian noise to the second image comprises: adding Gaussian noise to the second image according to a first formula; The first formula is: x i =x i-1 +g i-1 z; Where i = 1, 2, ..., N, N is a positive integer, g i-1 is the noise coefficient of the i-1th time Gaussian noise is added, z is Gaussian noise, x i is the second image after adding Gaussian noise for the i-1th time, x i-1 It is the second image after Gaussian noise is added for the i-2th time.

5. The method for generating PET images based on ultra-high field MRI according to claim 4, It is characterized in that The step of inputting the first image and the second image after adding Gaussian noise into a target neural network for deep learning to obtain a target joint score includes: The target joint score is obtained by solving the derivative of the logarithm of the joint probability distribution for the first image and the second image after Gaussian noise is added.

6. The method for generating PET images based on ultra-high field MRI according to claim 5, It is characterized in that The sampling of the initial image based on the target image in a backward process of a diffusion model according to the target joint score includes: Sampling the initial image according to a second formula; The second formula is: Where j = 1, 2, ..., N + 1, is the function corresponding to the target joint score obtained after the second image is processed for the N-j+1th time based on the first image, g j is the noise coefficient of adding Gaussian noise to the second image for the N-j+1th time, x′ j+1 is the initial image after the jth sampling, y′ is the target image, and f j (x′ j ) is the drift coefficient.

7. The method for generating PET images based on ultra-high field MRI according to claim 6, It is characterized in that The correcting the initial image based on the target image according to the annealed Langevin equation includes: Correcting the initial image according to a third formula; The third formula is: where ε=2(r||z|| 2 / ||s θ || 2 ), r is the parameter, x″ j+1 is the initial image after the j-th correction.

8. A PET image generation device based on ultra-high field MRI, It is characterized in that The device comprises: An image acquisition module, wherein the image acquisition module is used to acquire a target image, wherein the target image is an ultra-high field MRI image; A joint score acquisition module, the joint score acquisition module is used to acquire a target joint score, the target joint score is a derivative of a logarithm of a joint probability distribution of a first image and a second image in a forward process of a diffusion model, the first image and the second image being medical images taken for the same receptor; The image generation module processes the initial image multiple times based on the target joint score and the target image to obtain a target PET image, wherein the initial image is a pure Gaussian noise image, each processing includes sampling processing and correction processing, and each processing process includes: The initial image is sampled based on the target image in a backward process of a diffusion model according to the target joint score, and then the initial image is corrected based on the target image according to an annealed Langevin equation.

9. A terminal, It is characterized in that The terminal includes: a processor, a computer-readable storage medium communicatively connected to the processor, the computer-readable storage medium being suitable for storing a plurality of instructions, and the processor being suitable for calling the instructions in the storage medium to execute the steps of the method for generating PET images based on ultra-high field MRI as described in any one of claims 1 to 7.

10. A storage medium, It is characterized in that The storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps of the PET image generation method based on ultra-high field MRI as described in any one of claims 1-7.

Images