A three-dimensional brain blood volume image generation method, system, device and medium
By generating CBV images through diffusion models and consistency adversarial training, the problem of missing spatial structural information in the generation of three-dimensional cerebral blood volume images in existing technologies is solved, and high-quality CBV image generation is achieved, which is suitable for the diagnosis and treatment of gliomas.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NORTHWESTERN POLYTECHNICAL UNIV
- Filing Date
- 2026-05-13
- Publication Date
- 2026-07-14
AI Technical Summary
Existing methods lack spatial structural information when generating three-dimensional cerebral blood volume images, cannot obtain stable data, are highly dependent on operational experience, and are easily affected by patient movement.
A diffusion model combined with multimodal MRI conditional constraints and consistency adversarial training was adopted. CBV three-dimensional volume data was generated through a noise-adding network and a noise-reducing network. A discriminator was used for feature matching and adversarial loss function training to generate high-quality CBV images.
It enables the generation of high-quality CBV images without contrast agents, improves the spatial structural integrity and stability of the generated results, reduces examination risks and costs, and is suitable for CBV atlas acquisition in glioma diagnosis and treatment.
Smart Images
Figure CN122391456A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical image processing technology, and in particular to a method, system, device and medium for generating three-dimensional cerebral blood volume images. Background Technology
[0002] Magnetic resonance perfusion weighted imaging (MRPWI) is an important functional brain imaging technique that uses the injection of endogenous or exogenous contrast agents and the measurement of hemodynamic parameters to reflect tissue blood perfusion and microvascular-related functional information. Perfusion imaging can provide information on local tissue blood perfusion and hemodynamic changes, which is of great reference value for clinical diagnosis and treatment, and has been widely used in the pathological evaluation of the central nervous system.
[0003] In the diagnosis and treatment of brain tumors, especially gliomas, perfusion imaging can not only assist in tumor grading but also help locate more aggressive areas within the tumor to guide biopsies. For patients with treated high-grade gliomas, conventional magnetic resonance imaging (MRI) sequences have difficulty distinguishing tumor progression from the response to radiotherapy. Clinically, dynamic sensitivity contrast perfusion imaging (DSC)-MRI is often used to differentiate cerebral blood volume (CBV). Therefore, CBV atlases have become an important basis for relevant imaging monitoring.
[0004] However, perfusion imaging typically requires high-flow-rate contrast agent infusion, making it less adaptable to changes in the patient's physiological state. Furthermore, image post-processing is highly dependent on operator experience and easily influenced by the skill level of the technician, increasing the risk of contrast agent injection and post-processing failures. In addition, even slight patient movement can adversely affect post-processing results. These factors frequently lead to missing or unstable CBV (contrast volume image) atlases in real-world medical imaging scenarios.
[0005] Current generative artificial intelligence technologies offer potential solutions for missing modal image completion, and there has been some progress in the field of medical imaging in generating missing modalities using existing modalities. However, while existing methods have touched upon 3D medical image generation, most studies still focus on generation between conventional MRI modalities. In the research on generating 3D CBV images, which is of great value in the diagnosis and treatment of gliomas, this approach is prone to resulting in the loss of spatial structural information in the generated results, making it unsuitable for generating 3D cerebral blood volume images. Summary of the Invention
[0006] The purpose of this invention is to provide a method, system, device, and medium for generating three-dimensional cerebral blood volume images, which can solve the problem that existing methods lack spatial structure information in the generated results and cannot achieve the generation of three-dimensional cerebral blood volume images.
[0007] To address the aforementioned technical problems, embodiments of the present invention provide a method for generating three-dimensional cerebral blood volume images, comprising the following steps: MRI 3D volumetric data and real CBV 3D volumetric data of several objects were acquired respectively; Based on the diffusion model, a noise-adding network is used to progressively add noise to each real CBV three-dimensional volume data to obtain a noisy sample. Then, a denoising network is used to progressively denoise the joint input consisting of each noisy sample and the corresponding MRI three-dimensional volume data to obtain the generated CBV three-dimensional volume data. A denoising network is used as a generator to obtain all generated CBV 3D volume data obtained by the generator. A discriminator is used to judge each generated CBV 3D volume data according to the feature matching degree between the generated CBV 3D volume data and the corresponding real CBV 3D volume data. The discriminator is trained using an adversarial loss function built based on the discriminator's discrimination results, and the generator is trained using an adversarial consistency loss function built based on the discriminator's discrimination results and the degree of feature matching, so as to obtain a 3D CBV image generation model. The MRI three-dimensional volume data of the target object is input into the three-dimensional CBV image generation model to generate a three-dimensional CBV image of the target object.
[0008] Further, the step of progressively adding noise to each real CBV three-dimensional volume data using a noise-adding network to obtain a noisy sample, and progressively denoising the joint input consisting of each noisy sample and the corresponding MRI three-dimensional volume data using a denoising network, includes: Gaussian noise is gradually added to each real CBV 3D volume data according to a preset noise schedule using a noise-adding network to obtain noise-added samples at different time steps; The noisy sample at each time step is concatenated with the corresponding MRI 3D volume data in the channel dimension to obtain a joint input, which is then fed into the denoising network for progressive denoising.
[0009] Furthermore, the MRI three-dimensional volume data of each object includes at least one MRI modality of three-dimensional volume data; The step of concatenating the noisy samples at each time step with the corresponding MRI three-dimensional volume data in the channel dimension to obtain the joint input includes: Three-dimensional volume data of at least one MRI modality are randomly selected from MRI three-dimensional volume data and spliced in the channel dimension to form a conditional tensor; The noisy samples at each time step are concatenated with the corresponding conditional tensors along the channel dimension to obtain the joint input.
[0010] Furthermore, in each iteration of the training of the 3D CBV image generation model, each conditional tensor is formed by stitching together 3D volume data of at least one MRI modality randomly selected from the MRI 3D volume data.
[0011] Furthermore, the denoising network employs the DDIM accelerated sampling algorithm to progressively denoise the joint input consisting of each noisy sample and the corresponding MRI three-dimensional volume data.
[0012] Furthermore, after acquiring the MRI three-dimensional volume data and real CBV three-dimensional volume data of several objects respectively, the process also includes: All MRI 3D volume data and real CBV 3D volume data underwent the following preprocessing: orientation / voxel spacing uniformity, spatial alignment, intensity normalization, and abnormal intensity truncation.
[0013] Furthermore, the diffusion model adopts a three-dimensional U-Net network structure, and the discriminator adopts a three-dimensional convolutional network structure.
[0014] Embodiments of the present invention also provide a three-dimensional cerebral blood volume image generation system, comprising: The data acquisition module is used to acquire MRI three-dimensional volume data and real CBV three-dimensional volume data of several objects respectively; The model training module is used to progressively add noise to each real CBV 3D volume data based on the diffusion model through a noise-adding network to obtain noisy samples, and progressively denoise the joint input consisting of each noisy sample and the corresponding MRI 3D volume data through a denoising network to obtain generated CBV 3D volume data. A denoising network is used as a generator to obtain all generated CBV 3D volume data obtained by the generator. A discriminator is used to judge each generated CBV 3D volume data according to the feature matching degree between the generated CBV 3D volume data and the corresponding real CBV 3D volume data. The discriminator is trained using an adversarial loss function built based on the discriminator's discrimination results, and the generator is trained using an adversarial consistency loss function built based on the discriminator's discrimination results and the degree of feature matching, so as to obtain a 3D CBV image generation model. The image generation module is used to input the MRI three-dimensional volume data of the target object into the three-dimensional CBV image generation model to generate a three-dimensional CBV image of the target object.
[0015] Embodiments of the present invention also provide a computer device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described three-dimensional cerebral blood volume image generation method.
[0016] Embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described method for generating three-dimensional cerebral blood volume images.
[0017] The method for generating three-dimensional cerebral blood volume images provided by this invention has at least the following beneficial effects: This invention generates corresponding CBV 3D volume data using readily available MRI 3D volume data, addressing the common problem of missing or unstable CBV atlases in real-world medical imaging scenarios. Specifically, during the CBV 3D volume data generation process, a diffusion model learns the mapping between MRI 3D volume data and CBV 3D volume data, ensuring the integrity of the 3D spatial structure of the generated CBV image. Simultaneously, a discriminator constrains the CBV image generation process by evaluating the discrimination results of the diffusion model-generated CBV 3D volume data and the feature matching degree between the generated CBV 3D volume data and the corresponding real CBV 3D volume data, further improving the quality of the generated CBV image. Attached Figure Description
[0018] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:
[0019] Figure 1 A flowchart illustrating a method for generating three-dimensional cerebral blood volume images provided by the present invention; Figure 2 A schematic diagram of the network framework for a three-dimensional cerebral blood volume image generation method provided by the present invention. Detailed Implementation
[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.
[0021] Currently, most studies focus on the generation of CBV images between conventional MRI modalities, with limited research on brain perfusion MRI, particularly the generation of CBV maps, which is of significant value in the diagnosis and treatment assessment of gliomas. This invention proposes a method for generating CBV images based on a diffusion model combined with multimodal MRI conditional constraints and consistency adversarial training, effectively utilizing conventional MRI images to generate high-quality cerebral blood volume atlases.
[0022] The technical solutions provided by the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0023] One embodiment of the present invention relates to a method for generating three-dimensional cerebral blood volume images. The implementation details of the three-dimensional cerebral blood volume image generation method of this embodiment are described in detail below. The following content is only for the convenience of understanding and is not necessary for implementing this solution.
[0024] The specific process of the three-dimensional cerebral blood volume image generation method in this embodiment can be described as follows: Figure 1 As shown, it includes: Step 101: Obtain MRI 3D volumetric data and real CBV 3D volumetric data for several objects respectively; Specifically, after obtaining MRI three-dimensional volume data and real CBV three-dimensional volume data of several objects, the following preprocessing is performed on all MRI three-dimensional volume data and real CBV three-dimensional volume data: orientation / voxel spacing unification, spatial alignment, intensity normalization, and abnormal intensity truncation.
[0025] Step 102: Based on the diffusion model, noise is gradually added to each real CBV three-dimensional volume data through a noise-adding network to obtain a noise-adding sample. Then, noise is gradually denoised to the joint input consisting of each noise-adding sample and the corresponding MRI three-dimensional volume data through a denoising network to obtain the generated CBV three-dimensional volume data. Specifically, Gaussian noise is progressively added to each real CBV 3D volume data according to a preset noise schedule via a diffusion model's noise-adding network, resulting in noisy samples at different time steps. Then, the noisy samples at each time step are concatenated with the corresponding MRI 3D volume data along the channel dimension to obtain a joint input, which is then fed into a denoising network for progressive denoising. The diffusion model employs a 3D U-Net network structure or an equivalent denoising network structure based on a 3D encoder-decoder.
[0026] In one example, the denoising network uses the DDIM accelerated sampling algorithm to progressively denoise each noisy sample and the corresponding MRI 3D volume data as a joint input.
[0027] In a specific implementation, the MRI three-dimensional volume data of each object obtained in step 101 in this embodiment includes at least one MRI modality of three-dimensional volume data. By randomly selecting at least one MRI modality of three-dimensional volume data from the MRI three-dimensional volume data, and splicing them in the channel dimension to form a conditional tensor, the noisy sample of each time step and the corresponding conditional tensor are spliced in the channel dimension to obtain a joint input, which is then input into the denoising network for gradual denoising.
[0028] Step 103: Using a denoising network as a generator, obtain all generated CBV 3D volume data obtained through the generator, and use a discriminator to discriminate each generated CBV 3D volume data according to the feature matching degree between the generated CBV 3D volume data and the corresponding real CBV 3D volume data.
[0029] Step 104: Train the discriminator using an adversarial loss function constructed based on the discriminator's discrimination results, and train the generator using an adversarial consistency loss function constructed based on the discriminator's discrimination results and feature matching degree, to obtain a 3D CBV image generation model.
[0030] In one example, during each iteration of the 3D CBV image generation model training, each conditional tensor is formed by stitching together 3D volume data from at least one MRI modality that has been randomly reselected from MRI 3D volume data.
[0031] In addition, the discriminator in this embodiment adopts a three-dimensional convolutional network structure.
[0032] Step 105: Input the MRI three-dimensional volume data of the target object into the three-dimensional CBV image generation model to generate a three-dimensional CBV image of the target object.
[0033] Specifically, by using MRI three-dimensional volume data of several objects and real CBV three-dimensional volume data, and training the model using the above steps, the MRI three-dimensional volume data of the target object is input into the trained three-dimensional CBV image generation model to generate the CBV three-dimensional volume data of the target object, i.e., the three-dimensional CBV image.
[0034] In some embodiments, see Figure 2 The network framework shown illustrates the method for generating three-dimensional cerebral blood volume images according to the present invention. The specific implementation steps are as follows: 1. Data Acquisition and Preprocessing: Conventional MRI 3D volumetric data from the same subject, such as one or more of T1, T2, and T2-FLAIR, are acquired as conditional inputs. During the training phase, spatially registered real CBV 3D volumetric data are acquired as supervised targets. All volumetric data undergo preprocessing, preferably including: orientation / voxel spacing unification, spatial alignment, intensity normalization, and abnormal intensity truncation, to reduce the impact of differences in different devices and sequences on training.
[0035] 2. Noise addition process in the diffusion model: like Figure 2 As shown above, Gaussian noise is gradually added to the real CBV volume data during training according to a preset noise schedule, resulting in noisy samples at different time steps. This process is used to construct a denoising learning task, enabling the model to learn to gradually recover the real CBV volume data from the noisy samples given conditional information.
[0036] 3. Condition Construction and "Random Assembly": like Figure 2 As indicated by the labels "conventional MRI" and "randomly spliced," this embodiment uses conventional MRI as conditional constraint information during both training and inference. To improve the model's robustness to multimodal input missingness, modal differences, and redundant information, the following condition construction method is adopted: one or more modalities are randomly selected from the available set of conventional MRI modalities; the selected modalities are spliced along the channel dimension to form a conditional tensor; during training, the splicing combination can be randomly reselected for each iteration / each sample to enhance generalization ability.
[0037] 4. Denoising network and conditional fusion: like Figure 2 As shown in the green area in the middle, during training, the noisy sample is started and fused with the conditional tensor obtained in step 3 as input to the denoising network. First, the input is concatenated along the channel dimension to obtain the joint input. This joint input is then fed into a three-dimensional denoising network (the multi-layer encoder-decoder structure shown in the middle of the attached diagram, which could be a three-dimensional U-Net or an equivalent structure). The network outputs an estimate of the noise or residual, and based on this, performs a back-reasoning step to obtain the noisy sample from the previous step. This back-reasoning step is repeated to gradually obtain the generated CBV volume data, i.e. Figure 2 As shown in the cube on the right.
[0038] 5. Consistency adversarial constraints and discriminator training: like Figure 2 As shown on the right, the discriminator receives the real CBV and the generated CBV and outputs the discrimination result; the adversarial loss prompts the generator (denoising network) to generate more realistic CBV volume data while improving the discriminator's ability to distinguish. Meanwhile, the discriminator can employ a 3D convolutional structure and introduce feature consistency constraints at multi-scale feature levels, such as feature matching loss, to further stabilize training and improve generation quality. Ultimately, the generator is trained as a weighted combination of diffusion denoising loss and adversarial consistency loss; the discriminator is trained with an adversarial objective.
[0039] 6. Generation process of the reasoning stage: like Figure 2 As shown in the dashed box at the bottom, no real CBV is required during the inference phase. Initial noise volume data is obtained through random sampling; the subject's conventional MRI is randomly stitched together as in step 3 to obtain the conditional tensor; the stitched combination is input into the denoising network for step-by-step back-inference to obtain the final generated 3D CBV image. To improve inference efficiency, back-inference sampling can be accelerated using the DDIM class with fewer steps, shortening the generation time while maintaining quality.
[0040] This invention addresses the pain points of difficult, costly, and unstable CBV perfusion information acquisition in the clinical diagnosis of gliomas. It proposes a CBV generation scheme based on conventional multimodal MRI, utilizing three-dimensional conditional diffusion and introducing consistency adversarial constraints. This achieves three-dimensional CBV atlas reconstruction without contrast agents or perfusion scanning. The scheme enhances robustness to missing input modalities and cross-center differences through a randomized splicing conditional construction mechanism, and improves the detail accuracy and structural consistency of the generated results through adversarial consistency constraints. This provides more readily available functional imaging evidence for glioma grading assessment, treatment monitoring, and prognostic analysis, demonstrating significant application value in reducing examination risks and costs, improving image accessibility and standardization, and promoting the implementation of related intelligent diagnostic systems.
[0041] The steps of the various methods described above are only for clarity. In practice, they can be combined into one step or some steps can be split into multiple steps. As long as they include the same logical relationship, they are all within the protection scope of this invention. Adding insignificant modifications or introducing insignificant designs to the algorithm or process, without changing the core design of the algorithm and process, are also within the protection scope of this invention.
[0042] Another embodiment of the present invention relates to a three-dimensional cerebral blood volume image generation system. The implementation details of this embodiment's three-dimensional cerebral blood volume image generation system are described below. The following implementation details are provided for ease of understanding and are not essential for implementing this solution. The three-dimensional cerebral blood volume image generation system of this embodiment includes: The data acquisition module is used to acquire MRI three-dimensional volume data and real CBV three-dimensional volume data of several objects respectively; The model training module is used to progressively add noise to each real CBV 3D volume data based on the diffusion model through a noise-adding network to obtain noisy samples, and progressively denoise the joint input consisting of each noisy sample and the corresponding MRI 3D volume data through a denoising network to obtain generated CBV 3D volume data. A denoising network is used as a generator to obtain all generated CBV 3D volume data obtained by the generator. A discriminator is used to judge each generated CBV 3D volume data according to the feature matching degree between the generated CBV 3D volume data and the corresponding real CBV 3D volume data. The discriminator is trained using an adversarial loss function built based on the discriminator's discrimination results, and the generator is trained using an adversarial consistency loss function built based on the discriminator's discrimination results and the degree of feature matching, so as to obtain a 3D CBV image generation model. The image generation module is used to input the MRI three-dimensional volume data of the target object into the three-dimensional CBV image generation model to generate a three-dimensional CBV image of the target object.
[0043] It is not difficult to see that this embodiment is a system embodiment corresponding to the above method embodiments, and this embodiment can be implemented in conjunction with the above method embodiments. The relevant technical details and technical effects mentioned in the above embodiments are still valid in this embodiment, and will not be repeated here to reduce repetition. Accordingly, the relevant technical details mentioned in this embodiment can also be applied to the above embodiments.
[0044] It is worth mentioning that all modules involved in this embodiment are logical modules. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. Furthermore, to highlight the innovative aspects of this invention, this embodiment does not introduce units that are not closely related to solving the technical problem proposed by this invention; however, this does not mean that other units are absent from this embodiment.
[0045] Another embodiment of the present invention relates to a computer device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the three-dimensional cerebral blood volume image generation method of the above embodiments.
[0046] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and memories. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.
[0047] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory is used to store data used by the processor during operation.
[0048] Another embodiment of the present invention relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the method embodiments described above.
[0049] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0050] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing the present invention, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of the present invention.
Claims
1. A method for generating a three-dimensional cerebral blood volume image, characterized in that, The method includes: MRI 3D volumetric data and real CBV 3D volumetric data of several objects were acquired respectively; Based on the diffusion model, a noise-adding network is used to progressively add noise to each real CBV three-dimensional volume data to obtain a noisy sample. Then, a denoising network is used to progressively denoise the joint input consisting of each noisy sample and the corresponding MRI three-dimensional volume data to obtain the generated CBV three-dimensional volume data. A denoising network is used as a generator to obtain all generated CBV 3D volume data obtained by the generator. A discriminator is used to judge each generated CBV 3D volume data according to the feature matching degree between the generated CBV 3D volume data and the corresponding real CBV 3D volume data. The discriminator is trained using an adversarial loss function built based on the discriminator's discrimination results, and the generator is trained using an adversarial consistency loss function built based on the discriminator's discrimination results and the degree of feature matching, so as to obtain a 3D CBV image generation model. The MRI three-dimensional volume data of the target object is input into the three-dimensional CBV image generation model to generate a three-dimensional CBV image of the target object.
2. The method for generating a three-dimensional cerebral blood volume image according to claim 1, characterized in that, The process involves progressively adding noise to each real CBV 3D volume data using a noise-adding network to obtain noisy samples, and then progressively denoising the joint input consisting of each noisy sample and the corresponding MRI 3D volume data using a denoising network. This includes: Gaussian noise is gradually added to each real CBV 3D volume data according to a preset noise schedule using a noise-adding network to obtain noise-added samples at different time steps; The noisy sample at each time step is concatenated with the corresponding MRI 3D volume data in the channel dimension to obtain a joint input, which is then fed into the denoising network for progressive denoising.
3. The method for generating a three-dimensional cerebral blood volume image according to claim 2, characterized in that, The MRI three-dimensional volume data of each of the objects contains at least one MRI modality of three-dimensional volume data; The step of concatenating the noisy samples at each time step with the corresponding MRI three-dimensional volume data in the channel dimension to obtain the joint input includes: Three-dimensional volume data of at least one MRI modality are randomly selected from MRI three-dimensional volume data and spliced in the channel dimension to form a conditional tensor; The noisy samples at each time step are concatenated with the corresponding conditional tensors along the channel dimension to obtain the joint input.
4. The method for generating a three-dimensional cerebral blood volume image according to claim 3, characterized in that, In each iteration of the training of the 3D CBV image generation model, each conditional tensor is formed by stitching together 3D volume data of at least one MRI modality randomly selected from MRI 3D volume data.
5. The method for generating a three-dimensional cerebral blood volume image according to claim 1, characterized in that, The denoising network uses the DDIM accelerated sampling algorithm to progressively denoise each noisy sample and the corresponding MRI three-dimensional volume data as a joint input.
6. The method for generating a three-dimensional cerebral blood volume image according to claim 1, characterized in that, After acquiring the MRI three-dimensional volume data and real CBV three-dimensional volume data of several objects respectively, the method further includes: All MRI 3D volume data and real CBV 3D volume data underwent the following preprocessing: orientation / voxel spacing uniformity, spatial alignment, intensity normalization, and abnormal intensity truncation.
7. The method for generating a three-dimensional cerebral blood volume image according to claim 1, characterized in that, The diffusion model adopts a three-dimensional U-Net network structure, and the discriminator adopts a three-dimensional convolutional network structure.
8. A three-dimensional cerebral blood volume image generation system, characterized in that, The system includes: The data acquisition module is used to acquire MRI three-dimensional volume data and real CBV three-dimensional volume data of several objects respectively; The model training module is used to progressively add noise to each real CBV 3D volume data based on the diffusion model through a noise-adding network to obtain noisy samples, and progressively denoise the joint input consisting of each noisy sample and the corresponding MRI 3D volume data through a denoising network to obtain generated CBV 3D volume data. A denoising network is used as a generator to obtain all generated CBV 3D volume data obtained by the generator. A discriminator is used to judge each generated CBV 3D volume data according to the feature matching degree between the generated CBV 3D volume data and the corresponding real CBV 3D volume data. The discriminator is trained using an adversarial loss function built based on the discriminator's discrimination results, and the generator is trained using an adversarial consistency loss function built based on the discriminator's discrimination results and the degree of feature matching, so as to obtain a 3D CBV image generation model. The image generation module is used to input the MRI three-dimensional volume data of the target object into the three-dimensional CBV image generation model to generate a three-dimensional CBV image of the target object.
9. A computer device, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the three-dimensional cerebral blood volume image generation method as described in any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for generating three-dimensional cerebral blood volume images as described in any one of claims 1 to 7.