MRI image reconstruction method, device, electronic device and storage medium

Abstract

The present application relates to an MRI image reconstruction method, device, electronic device, and storage medium, wherein the MRI image reconstruction method includes: obtaining multi-contrast MRI sample data, including target modality and auxiliary modality high-resolution images; performing frequency domain conversion on the target modality high-resolution image and truncating high-frequency components to generate low-resolution frequency domain data; then performing spatial domain conversion on the target modality high-resolution image to generate a spatial domain low-resolution image; inputting the low-resolution frequency domain data and the spatial domain low-resolution image into a diffusion model, using the diffusion model to extract features of the two types of low-resolution data in the frequency domain and spatial domain respectively to generate current time step features; and using a structural attention mechanism to perform feature fusion on the current time step features and the features of the auxiliary modality high-resolution image to obtain a reconstructed image of the target modality. Through this application, the difficult problem of balancing spectral fidelity and detail enhancement in traditional MRI super-resolution methods is solved, significantly improving reconstruction efficiency and image quality.

Description

Technical Field

[0001] The present application relates to the field of medical image processing, and in particular to an MRI image reconstruction method, device, electronic device, and storage medium. Background Art

[0002] Magnetic resonance imaging (MRI) is a core tool for clinical diagnosis, and its resolution directly affects the accuracy of lesion detection. However, due to limitations in hardware cost, scanning time, and patient tolerance, low-resolution (LR) MRI images are prevalent, making it difficult to identify tiny lesions (such as early-stage tumors and microbleeds).

[0003] Current mainstream technologies fall into three categories: frequency-domain reconstruction methods, spatial-domain deep learning models, and diffusion models. However, all have significant limitations. First, frequency-domain methods (such as k-space reconstruction) directly supplement high-frequency signals in the frequency domain through techniques such as compressed sensing. While these methods maintain spectral fidelity, they can easily lose spatial texture details (such as vascular branches and tissue edges), resulting in visual blurriness in the reconstructed image and making it difficult to meet clinical diagnostic requirements. Second, spatial-domain deep learning models (such as U-Net) rely on end-to-end learning in the image domain and can effectively enhance visual detail. However, they lack the ability to model the physical constraints of imaging, potentially leading to anatomical distortion (such as organ shape shifts), compromising diagnostic reliability. Third, direct application of diffusion models to MRI involves progressively denoising random noise to generate images. This misalignment with the inherent nature of MRI degradation (loss of low-frequency information), forcing the model to start from pure noise, requiring numerous iterations and time-consuming inference. Furthermore, because MRI degradation is not global noise contamination, blind denoising can easily introduce artifacts (such as false lesions). In addition, existing dual-domain (frequency domain + spatial domain) fusion mostly adopts static strategies and cannot dynamically exchange cross-domain high-frequency residuals, resulting in error accumulation.

[0004] Currently, no effective solution has been proposed to the difficult problem of balancing spectral fidelity and detail enhancement in traditional MRI super-resolution methods. Summary of the Invention

[0005] The embodiments of the present application provide an MRI image reconstruction method, apparatus, electronic device, and storage medium to at least solve the difficult problem of balancing spectral fidelity and detail enhancement in traditional MRI super-resolution methods.

[0006] In a first aspect, an embodiment of the present application provides an MRI image reconstruction method, comprising:

[0007] Acquire multi-contrast MRI sample data, including target modality high-resolution images and auxiliary modality high-resolution images;

[0008] Performing frequency domain conversion on the target modality high-resolution image to obtain initial frequency domain data, truncating high-frequency components of the initial frequency domain data to generate low-resolution frequency domain data; performing spatial domain conversion on the low-resolution frequency domain data to generate a spatial domain low-resolution image;

[0009] The low-resolution frequency domain data and the spatial domain low-resolution image are input into a diffusion model. The diffusion model is used to extract features of the low-resolution frequency domain data and the spatial domain low-resolution image in the frequency domain and the spatial domain, respectively, to generate current time step features. The structural attention mechanism is used to perform feature fusion on the current time step features and the features of the auxiliary modality high-resolution image to obtain a reconstructed image of the target modality.

[0010] In some embodiments, truncating high-frequency components of the initial frequency domain data to generate low-resolution frequency domain data includes:

[0011] Obtain a preset mask that blocks high-frequency components from the edge to the center according to a preset step size. Based on the initial frequency domain data, reduce the mask in the frequency center area according to the preset step size, cut off the high-frequency components, and generate the low-resolution frequency domain data at different time steps.

[0012] In some embodiments, the extracting features of the low-resolution frequency domain data and the spatial domain low-resolution image in the frequency domain and the spatial domain, respectively, to generate current time step features; and using a structural attention mechanism to fuse the current time step features with features of the auxiliary modality high-resolution image to obtain a reconstructed image of the target modality includes:

[0013] The current time step feature includes a frequency domain time step feature and a spatial domain time step feature;

[0014] In the frequency domain, encoder feature extraction is performed on the low-resolution frequency domain data to generate the frequency domain time step feature; the frequency domain time step feature is fused with the frequency domain feature of the auxiliary modality high-resolution image using the structural attention mechanism to generate a high-frequency missing distribution feature;

[0015] In the spatial domain, encoder features are extracted from the spatial domain low-resolution image to generate the spatial domain time step features; the spatial domain time step features are fused with the spatial domain features of the auxiliary modality high-resolution image using the structural attention mechanism to generate local detail features;

[0016] A reconstructed image of the target modality is acquired based on the high-frequency missing distribution feature and the local detail feature.

[0017] In some embodiments, the structural attention mechanism dynamically fuses the current time step features with the features of the auxiliary modality high-resolution image through attention weights.

[0018] In some embodiments, the method further comprises:

[0019] Acquire training MRI sample data, including high-resolution images of the training target modality and high-resolution images of the training auxiliary modality;

[0020] Performing frequency domain conversion on the training target modality high-resolution image to obtain training initial frequency domain data, truncating high-frequency components of the training initial frequency domain data to generate training low-resolution frequency domain data; performing spatial domain conversion on the training low-resolution frequency domain data to generate a training spatial domain low-resolution image;

[0021] Inputting the training low-resolution frequency domain data and the training spatial domain low-resolution image into an initial diffusion model, using the initial diffusion model, performing feature extraction on the training low-resolution frequency domain data and the training spatial domain low-resolution image in the frequency domain and the spatial domain, respectively, to generate training current time step features; and using a structured attention mechanism, performing feature fusion on the training current time step features and the features of the training auxiliary modality high-resolution image to reconstruct a training reconstructed image of the target modality;

[0022] The training reconstructed image includes frequency domain reconstructed data and spatial domain reconstructed data;

[0023] Constructing a dual-domain data consistency loss function and a reconstruction loss function; wherein the dual-domain data consistency loss function includes a frequency domain consistency loss function and a spatial domain consistency loss function;

[0024] Calculating a frequency domain consistency loss value, a spatial domain consistency loss value, and a reconstruction loss value based on the dual-domain data consistency loss function and the reconstruction loss function;

[0025] The frequency domain consistency loss value, the spatial domain consistency loss value and the reconstruction loss value are weightedly fused to obtain a total loss function value; based on the total loss function value, the parameters of the initial diffusion model are iteratively optimized to generate a trained diffusion model.

[0026] In some embodiments, the calculating the frequency domain consistency loss value, the spatial domain consistency loss value, and the reconstruction loss value based on the dual-domain data consistency loss function and the reconstruction loss function includes:

[0027] Based on the frequency domain consistency loss function, calculating the difference between the frequency domain reconstructed image features and the corresponding frequency domain features of the training target modality high-resolution image to generate a frequency domain consistency loss value;

[0028] Based on the spatial consistency loss function, calculating the difference between the spatial reconstructed image features and the corresponding spatial features of the training target modality high-resolution image to generate a spatial consistency loss value;

[0029] Based on the reconstruction loss function, the difference between the features of the training reconstructed image and the corresponding high-resolution image of the training target modality is calculated to generate a reconstruction loss value.

[0030] In some embodiments, the weighted fusion of the frequency domain consistency loss value, the spatial domain consistency loss value, and the reconstruction loss value includes:

[0031] When the proportion of low-frequency components in the training low-resolution frequency domain data is greater than that of high-frequency components, the weight value of the frequency domain consistency loss value is greater than the weight value of the spatial domain consistency loss value;

[0032] When the proportion of low-frequency components in the training low-resolution frequency domain data is not greater than that of high-frequency components, the weight value of the spatial domain consistency loss value is greater than the weight value of the frequency domain consistency loss value;

[0033] Based on the weight value of the spatial domain consistency loss value, the weight value of the frequency domain consistency loss value and the weight value of the reconstruction loss value, the frequency domain consistency loss value, the spatial domain consistency loss value and the reconstruction loss value are weightedly fused.

[0034] In a second aspect, an embodiment of the present application provides an MRI image reconstruction device, comprising:

[0035] An image acquisition module, used to acquire multi-contrast MRI sample data, including target modality high-resolution images and auxiliary modality high-resolution images;

[0036] a data processing module configured to perform frequency domain conversion on the target modality high-resolution image to obtain initial frequency domain data, truncate high-frequency components of the initial frequency domain data to generate low-resolution frequency domain data, and perform spatial domain conversion on the low-resolution frequency domain data to generate a spatial domain low-resolution image;

[0037] The image reconstruction module is used to input the low-resolution frequency domain data and the spatial domain low-resolution image into a diffusion model, and use the diffusion model to extract features of the low-resolution frequency domain data and the spatial domain low-resolution image in the frequency domain and spatial domain respectively to generate current time step features; and use the structural attention mechanism to feature fuse the current time step features with the features of the auxiliary modality high-resolution image to obtain a reconstructed image of the target modality.

[0038] In a third aspect, an embodiment of the present application provides an electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, the MRI image reconstruction method as described in the first aspect above is implemented.

[0039] In a fourth aspect, an embodiment of the present application provides a storage medium on which a computer program is stored, and when the program is executed by a processor, the MRI image reconstruction method as described in the first aspect above is implemented.

[0040] Compared with related technologies, the embodiments of the present application provide an MRI image reconstruction method, device, electronic device and storage medium. By acquiring multi-contrast MRI data (including target and auxiliary modality high-resolution images), the target modality image is converted to the frequency domain and the high frequency is truncated to generate low-resolution frequency domain data, and then converted back to the spatial domain to obtain a low-resolution image; the two types of low-resolution data are input into a diffusion model, features are extracted in the frequency domain and spatial domain respectively, and the auxiliary modality features are fused with the structural attention mechanism to finally reconstruct the target modality high-resolution image, which solves the difficult problem of balancing spectral fidelity and detail enhancement in traditional MRI super-resolution methods and significantly improves reconstruction efficiency and image quality.

[0041] The details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0043] Figure 1 1 is a hardware structure block diagram of a terminal of the MRI image reconstruction method according to an embodiment of the present invention;

[0044] Figure 2 is a flowchart of an MRI image reconstruction method according to an embodiment of the present application;

[0045] Figure 3 is a preferred flow chart of the MRI image reconstruction method according to the preferred embodiment of the present application;

[0046] Figure 4 1 is a framework diagram of a dual-domain cooperative diffusion model of an MRI image reconstruction method according to a preferred embodiment of the present application;

[0047] Figure 5 It is a structural block diagram of an MRI image reconstruction device according to an embodiment of the present application. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is described and illustrated below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. Based on the embodiments provided in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application. In addition, it can also be understood that although the efforts made in this development process may be complex and lengthy, for ordinary technicians in the field related to the contents disclosed in the present application, some changes such as design, manufacturing or production based on the technical contents disclosed in the present application are only conventional technical means and should not be understood as the contents disclosed in the present application being insufficient.

[0049] References to "embodiments" in this application mean that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it refer to independent or alternative embodiments that are mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described in this application may be combined with other embodiments unless there is a conflict.

[0050] Unless otherwise defined, technical or scientific terms used herein shall have the ordinary meaning as understood by persons of ordinary skill in the art to which this application belongs. The terms "a," "an," "an," "the," and similar expressions used herein do not denote limitations on quantity and may refer to either the singular or the plural. The terms "comprise," "include," "have," and any variations thereof, used herein, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or modules (units) is not limited to the listed steps or units but may also include steps or units not listed, or may include other steps or units inherent to the process, method, product, or apparatus. The terms "connected," "connected," "coupled," and similar expressions used herein are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. As used herein, "plurality" means greater than or equal to two. "And / or" describes an association between associated objects, indicating that three possible relationships exist. For example, "A and / or B" may mean: A exists alone; A and B exist simultaneously; or B exists alone. The terms "first", "second", "third" and the like involved in this application are merely used to distinguish similar objects and do not represent a specific ordering of the objects.

[0051] The method embodiment provided in this embodiment can be executed in a terminal, a computer or a similar computing device. Taking running on a terminal as an example, Figure 1 FIG. 1 is a block diagram of the hardware structure of the terminal of the MRI image reconstruction method according to an embodiment of the present invention. Figure 1 As shown, the terminal may include one or more ( Figure 1 Only one is shown) a processor 102 (the processor 102 may include but is not limited to a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data. Optionally, the terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It will be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above terminal. Figure 1 More or fewer components than shown, or with Figure 1 Different configurations shown.

[0052] The memory 104 can be used to store computer programs, such as software programs and modules of application software, such as the computer program corresponding to the MRI image reconstruction method in the embodiment of the present invention. The processor 102 executes the computer program stored in the memory 104 to perform various functional applications and data processing, thereby implementing the above-mentioned method. The memory 104 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0053] Transmission device 106 is used to receive or transmit data via a network. A specific example of the aforementioned network may include a wireless network provided by the terminal's communications provider. In one embodiment, transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0054] This embodiment provides an MRI image reconstruction method. Figure 2 is a flow chart of the MRI image reconstruction method according to an embodiment of the present application, such as Figure 2 As shown, the process includes the following steps:

[0055] Step S201 , acquiring multi-contrast MRI sample data, including a target modality high-resolution image and an auxiliary modality high-resolution image;

[0056] Multi-contrast MRI sample data is collected using medical imaging equipment (such as MRI scanners). The multi-contrast MRI sample data includes high-resolution images of the target modality (such as T2) and high-resolution images of the auxiliary modality (such as T1). The target modality high-resolution images are three-dimensional (3D) T2 high-resolution images (HR), which display tissue water content and highlight pathological features (such as tumors and edema); the auxiliary modality high-resolution images are three-dimensional (3D) T1 high-resolution images (HR), which clearly display anatomical structures (such as the boundaries between gray matter and white matter). The collected 3D MRI sample data is sliced ​​and converted into 2D data along the depth axis to adapt to the model input format. Preprocessing operations such as brightness normalization or Z-score standardization are performed on the images to eliminate device differences and noise interference. This step acquires multi-contrast MRI data to fuse complementary information from different modalities. MRI images of different modalities provide rich feature information. For example, T1 and T2 modalities may reveal different tissue characteristics. Combining this information can improve the detail and accuracy of the reconstructed image. High-quality multimodal MRI reconstructed images can provide more accurate information for clinical diagnosis, helping doctors make more accurate diagnostic and treatment decisions.

[0057] Step S202: performing frequency domain conversion on the target modality high-resolution image to obtain initial frequency domain data, truncating high-frequency components of the initial frequency domain data to generate low-resolution frequency domain data; performing spatial domain conversion on the low-resolution frequency domain data to generate a spatial domain low-resolution image;

[0058] The target modality's high-resolution image (e.g., T2 HR image) undergoes a frequency domain transformation (i.e., Fourier transform) to convert it from the spatial domain to the frequency domain, obtaining initial frequency domain data (k-space data). In the frequency domain, a mask can be defined that gradually blocks high-frequency components from the edge toward the center, simulating the k-space truncation degradation process. This mask gradually shrinks with increasing time steps, gradually truncating high-frequency components and generating low-resolution frequency domain data. This step simulates the information loss experienced during low-resolution MRI image acquisition. The resulting low-resolution frequency domain data undergoes a spatial domain transformation (i.e., inverse Fourier transform) to convert it back from the frequency domain to the spatial domain, generating a low-resolution spatial domain image. This step avoids the interference of invalid noise by truncating the high-frequency components and using low-frequency information as the diffusion starting point, which is more in line with the nature of MRI image degradation, because the degradation of MRI images is mainly due to the loss of low-frequency information rather than global noise pollution; by performing degradation processing in the frequency domain, the spectral fidelity can be maintained, which means that during the degradation process, the low-frequency components of the image are well preserved, which is helpful for the recovery of frequency domain information in the subsequent reconstruction process; since frequency domain processing can maintain spectral information, while spatial domain processing focuses more on the recovery of image details, compared with methods that rely only on frequency domain or spatial domain, the conversion processing of frequency domain and spatial domain can reduce the loss of spatial texture details to a certain extent. The generated low-resolution frequency domain data and spatial domain low-resolution images provide convenience for the subsequent dual-domain collaborative restoration process. These data can be used as input of the diffusion model for subsequent feature extraction and reconstruction processes.

[0059] In step S203, the low-resolution frequency domain data and the low-resolution spatial domain image are input into the diffusion model. The diffusion model is used to extract features of the low-resolution frequency domain data and the low-resolution spatial domain image in the frequency domain and the spatial domain, respectively, to generate the current time step features. The structural attention mechanism is used to fuse the current time step features with the features of the auxiliary modality high-resolution image to obtain a reconstructed image of the target modality.

[0060] The low-resolution frequency domain data and spatial domain low-resolution images generated in the above steps are fed into the diffusion model as input. Feature extraction is performed on both the low-resolution frequency domain data and the spatial domain low-resolution images. Frequency domain feature extraction focuses on capturing the image's spectral characteristics (such as phase distribution and low-frequency contours), while spatial domain feature extraction focuses on capturing the image's spatial structure and details (such as texture features such as edges and lesions). A specific encoder network structure generates feature representations for the current time step in both the frequency and spatial domains. A structured attention mechanism (SP-former) is then used to fuse the features extracted in the frequency and spatial domains at the current time step with features from an auxiliary modality high-resolution image (such as a T1 HR image). The structured attention mechanism adaptively learns correlations between different features and weights the features based on these correlations, generating a richer and more accurate feature representation. Based on the fused feature representations, a decoder network generates a reconstructed image of the target modality (such as a T2). This reconstructed image exhibits significant improvements in detail fidelity and reconstruction efficiency, providing high-quality clinical MRI imaging support.

[0061] Through the above steps, the present application uses low-resolution images as the diffusion starting point, and through dynamic collaborative optimization of k-space spectral fidelity and spatial details, it not only preserves the physical fidelity of the anatomical structure, but also effectively restores high-frequency details (such as vascular texture); at the same time, compared with the defect of anatomical position offset caused by the lack of frequency domain constraints in the spatial domain model in traditional methods, the present application uses multimodal feature fusion (such as combining T1 and T2 modalities) and structural attention mechanism to enhance visual details in the spatial domain while ensuring the accuracy of the anatomical structure through frequency domain consistency loss; in addition, in response to the problem that the traditional diffusion model starts reverse denoising from random noise and is essentially mismatched with MRI degradation, is inefficient and easily introduces artifacts, the present application directly uses low-frequency information as the diffusion starting point, reduces invalid noise interference, avoids a large number of iterative steps for restoring contours, thereby significantly improving reconstruction efficiency and image quality, solving the difficult problem of balancing spectral fidelity and detail enhancement in traditional MRI super-resolution methods, and achieving a perfect balance between spectral fidelity and detail enhancement.

[0062] In some embodiments, truncating high-frequency components of initial frequency domain data to generate low-resolution frequency domain data includes:

[0063] Obtain a preset mask that blocks high-frequency components from the edge to the center according to a preset step size. Based on the initial frequency domain data, reduce the mask of the frequency center area according to the preset step size, cut off the high-frequency components, and generate low-resolution frequency domain data at different time steps.

[0064] A mask is defined that blocks high-frequency components at a preset step size from the edge toward the center, simulating the gradual loss of high-frequency information in k-space (frequency domain). Based on the initial frequency domain data, the high-frequency components are truncated by gradually reducing the mask in the frequency center region. During this process, the pixel center coordinates of the mask remain unchanged, while the parameter controlling the side length of the retained k-space center region (i.e., the low-frequency components) gradually decreases with the number of steps. At each step, the initial frequency domain data is processed using the current mask to generate low-resolution frequency domain data for that time step. By gradually reducing the mask, a series of low-resolution frequency domain data at different time steps can be generated, reflecting the gradual loss of high-frequency information.

[0065] This embodiment truncates high-frequency components by gradually reducing the mask in the frequency center region. This accurately simulates the degradation process of MRI images during low-resolution acquisition, thus better aligning with the nature of MRI image degradation. This is because MRI image degradation is primarily due to the loss of low-frequency information (which is more significant than the loss of high-frequency information) rather than global noise contamination. By generating a series of low-resolution frequency domain data at different time steps, a diverse set of samples is provided for model training. These data cover varying degrees of degradation from high resolution to low resolution, helping the model learn more robust feature representations and improving the quality of reconstructed images.

[0066] In some embodiments, feature extraction is performed on low-resolution frequency domain data and spatial domain low-resolution images in the frequency domain and spatial domain, respectively, to generate current time step features; and a structural attention mechanism is used to perform feature fusion on the current time step features and features of the auxiliary modality high-resolution image to obtain a reconstructed image of the target modality, including:

[0067] The current time step features include frequency domain time step features and spatial domain time step features;

[0068] In the frequency domain, encoder features are extracted from low-resolution frequency domain data to generate frequency domain time step features. The frequency domain time step features are fused with the frequency domain features of the auxiliary modality high-resolution image using the structural attention mechanism to generate high-frequency missing distribution features.

[0069] In the spatial domain, encoder features are extracted from the spatial domain low-resolution image to generate spatial domain time step features. The structural attention mechanism is used to fuse the spatial domain time step features with the spatial domain features of the auxiliary modality high-resolution image to generate local detail features.

[0070] Based on the high-frequency missing distribution characteristics and local detail features, a reconstructed image of the target modality is obtained.

[0071] In the frequency domain, encoder features are extracted from the low-resolution frequency domain data to generate frequency-domain time-step features that capture low-frequency information and high-frequency loss patterns in the frequency domain. Using a structural attention mechanism, these frequency-domain time-step features are fused with the frequency-domain features of the auxiliary modality's high-resolution image to generate high-frequency loss distribution features. This structural attention mechanism adaptively learns the correlations between different features and weights the features based on these correlations, thereby highlighting regions with high-frequency loss. In the spatial domain, encoder features are extracted from the spatial domain's low-resolution image to generate spatial-domain time-step features that capture the image's spatial structure and local details. Similarly, using the structural attention mechanism, these spatial-domain time-step features are fused with the spatial-domain features of the auxiliary modality's high-resolution image to generate local detail features. This structural attention mechanism enhances the ability to capture and restore local image details. Based on the high-frequency loss distribution features and local detail features, the decoder network generates a reconstructed image of the target modality. These features provide the decoder with rich information, helping to preserve both spectral fidelity and spatial detail during reconstruction.

[0072] This embodiment fully utilizes the information complementarity of different domains by extracting and fusing features in the frequency domain and spatial domain respectively. Frequency domain features provide distribution information of high-frequency missing images, while spatial domain features provide rich local details. The combination of the two can significantly improve the quality of reconstructed images. The application of the structural attention mechanism in frequency domain feature fusion helps to more accurately identify and restore high-frequency missing areas, significantly improving the clarity and accuracy of high-frequency details (such as vascular texture) in the reconstructed image. In spatial domain feature fusion, the structural attention mechanism enhances the ability to capture and restore local details, allowing the reconstructed image to better present local details while maintaining the overall structure, thereby improving the visual quality of the image. By combining high-frequency missing distribution features and local detail features, the decoder can generate more accurate and clearer reconstructed images. This method achieves a good balance between spectral fidelity and spatial detail enhancement, providing more reliable MRI imaging support for clinical practice.

[0073] In some of these embodiments, the structural attention mechanism dynamically fuses the features of the current time step with the features of the auxiliary modality high-resolution image through attention weights.

[0074] The current time step features include frequency-domain and spatial-domain features, extracted from low-resolution frequency-domain data and spatial-domain low-resolution images using frequency-domain and spatial-domain encoders, respectively. Frequency-domain and spatial-domain features are extracted from the auxiliary modality high-resolution image using corresponding encoders. The structured attention mechanism calculates the similarity or correlation between the current time step features and the auxiliary modality high-resolution image features to obtain attention weights. The attention weights reflect the degree of correlation between different features and determine the contribution of each feature in the fusion process. The specific calculation process may involve operations such as dot products, weighted summation, and normalization of feature vectors to generate an attention weight matrix. Using the calculated attention weights, the current time step features and the auxiliary modality high-resolution image features are weightedly fused. In this process, features with higher attention weights account for a larger proportion of the fusion result, thus achieving dynamic feature selection and fusion. The fused features contain both the reconstruction information of the current time step and the beneficial features of the auxiliary modality high-resolution image, helping to improve the quality of the reconstructed image.

[0075] In this embodiment, the structural attention mechanism dynamically adjusts the weight distribution in the feature fusion process, and can adaptively adjust the contribution of each feature according to the correlation between the current time step feature and the auxiliary modality high-resolution image feature. This adaptive feature fusion method helps to better utilize the information of the auxiliary modality and improve the accuracy of the reconstructed image. At the same time, it also makes the model robust to noise and changes in the input data. Even when the MRI image input is of different modalities or different quality, the model can generate relatively stable reconstruction results.

[0076] In some embodiments, the method further comprises:

[0077] Acquire training MRI sample data, including high-resolution images of the training target modality and high-resolution images of the training auxiliary modality;

[0078] Perform frequency domain conversion on the high-resolution image of the training target modality to obtain the initial frequency domain data for training, truncate the high-frequency components of the initial frequency domain data for training to generate low-resolution frequency domain data for training; perform spatial domain conversion on the low-resolution frequency domain data for training to generate low-resolution images in the spatial domain for training;

[0079] The training low-resolution frequency domain data and the training spatial domain low-resolution images are input into the initial diffusion model. The initial diffusion model is used to extract features of the training low-resolution frequency domain data and the training spatial domain low-resolution images in the frequency domain and spatial domain, respectively, to generate the features of the training current time step. The structural attention mechanism is then used to fuse the features of the training current time step with the features of the training auxiliary modality high-resolution images to reconstruct the training reconstructed image of the target modality.

[0080] The training reconstructed image includes frequency domain reconstruction data and spatial domain reconstruction data;

[0081] Constructing a dual-domain data consistency loss function and a reconstruction loss function; wherein the dual-domain data consistency loss function includes a frequency domain consistency loss function and a spatial domain consistency loss function;

[0082] Based on the dual-domain data consistency loss function and the reconstruction loss function, the frequency domain consistency loss value, the spatial domain consistency loss value and the reconstruction loss value are calculated;

[0083] The frequency domain consistency loss value, spatial domain consistency loss value and reconstruction loss value are weighted and fused to obtain the total loss function value; based on the total loss function value, the parameters of the initial diffusion model are iteratively optimized to generate a trained diffusion model.

[0084] Training MRI sample data, including high-resolution images of the training target modality and high-resolution images of the training auxiliary modality, is obtained, providing the necessary input and reference standards for model training. The high-resolution images of the training target modality are frequency-domain transformed to obtain initial training frequency-domain data. The high-frequency components of the initial training frequency-domain data are truncated to generate low-resolution training frequency-domain data. This step simulates the frequency-domain degradation of MRI images during low-resolution acquisition. Spatial-domain transformation is performed on the training low-resolution frequency-domain data to generate low-resolution training spatial-domain images, providing a foundation for subsequent spatial-domain feature extraction. The training low-resolution frequency-domain data and the training spatial-domain low-resolution images are input into the initial diffusion model. Within the diffusion model, frequency-domain and spatial-domain feature extraction is performed on the training low-resolution frequency-domain data and the training spatial-domain low-resolution images, respectively, to generate features for the current training time step. Using a structural attention mechanism, the features of the current training time step are fused with the features of the high-resolution images of the training auxiliary modality to reconstruct the training reconstructed image of the target modality.

[0085] The training reconstructed image includes frequency domain reconstruction data and spatial domain reconstruction data, corresponding to the reconstruction results in the frequency domain and spatial domain, respectively. A dual-domain data consistency loss function and a reconstruction loss function are constructed. The dual-domain data consistency loss function includes a frequency domain consistency loss function and a spatial domain consistency loss function, which are used to measure the consistency between the frequency domain and spatial domain reconstructed data and the original high-resolution data, respectively. The reconstruction loss function is used to measure the difference between the reconstructed image and the true high-resolution image. Based on the dual-domain data consistency loss function and the reconstruction loss function, the frequency domain consistency loss value, the spatial domain consistency loss value, and the reconstruction loss value are calculated. These loss values ​​are weighted and fused to obtain the total loss function value. By adjusting the weights, the impact of different loss terms on model training can be balanced. Based on the total loss function value, an optimization algorithm (such as gradient descent) is used to iteratively optimize the parameters of the initial diffusion model to generate a trained diffusion model.

[0086] In this embodiment, the dual-domain data consistency loss function and the reconstruction loss function are used in combination, so that the model can simultaneously focus on the consistency of the frequency domain and the spatial domain, as well as the difference between the reconstructed image and the real image during the training process, which helps to improve the quality of the reconstructed image and make it closer to the real high-resolution image; by weighted fusion of different loss terms, the model can better balance the performance requirements of different aspects, thereby enhancing the robustness to input data noise and changes, and even when faced with MRI image inputs of different qualities, the model can generate relatively stable reconstruction results; the process of iterative optimization of model parameters enables the model to gradually learn more accurate feature representations and reconstruction strategies. By continuously adjusting the parameters to minimize the total loss function value, the model can gradually approach the optimal solution and improve the accuracy and clarity of the reconstructed image. The trained diffusion model can provide high-quality MRI image reconstruction results for clinical practice, helping doctors make more accurate diagnosis and treatment decisions. By improving the quality and robustness of the reconstructed image, the model has higher practical value and reliability in clinical applications.

[0087] In some embodiments, calculating a frequency domain consistency loss value, a spatial domain consistency loss value, and a reconstruction loss value based on a dual-domain data consistency loss function and a reconstruction loss function includes:

[0088] Based on the frequency domain consistency loss function, the difference between the frequency domain reconstructed image features and the corresponding frequency domain features of the training target modality high-resolution image is calculated to generate a frequency domain consistency loss value;

[0089] Based on the spatial consistency loss function, the difference between the spatial features of the reconstructed image and the corresponding spatial features of the training target modality high-resolution image is calculated to generate a spatial consistency loss value;

[0090] Based on the reconstruction loss function, the difference in features between the training reconstructed image and the corresponding training target modality high-resolution image is calculated to generate a reconstruction loss value.

[0091] The difference between the frequency-domain reconstructed image features and the frequency-domain features of the corresponding training target modality high-resolution image is calculated. A frequency-domain consistency loss function (such as the mean squared error loss function) is used to measure this difference, generating a frequency-domain consistency loss value. This ensures that the spectral characteristics of the reconstructed frequency-domain image are consistent with those of the true high-resolution image. The difference between the spatial-domain reconstructed image features and the spatial-domain features of the corresponding training target modality high-resolution image is calculated. A spatial-domain consistency loss function (such as the structural similarity loss function) is used to measure this difference, generating a spatial-domain consistency loss value. This ensures that the spatial structure and details of the reconstructed spatial-domain image are consistent with those of the true high-resolution image. The difference between the overall features of the training reconstructed image (including frequency-domain and spatial-domain reconstruction data) and the corresponding training target modality high-resolution image is calculated. A reconstruction loss function (such as the L1 loss function or the L2 loss function) is used to measure this difference, generating a reconstruction loss value. This ensures that the overall visual quality of the reconstructed image is consistent with that of the true high-resolution image.

[0092] In this embodiment, the calculation of the frequency domain consistency loss value ensures the accuracy and consistency of the reconstructed image in the frequency domain, helps to retain and restore the high-frequency details of the image, such as vascular texture, thereby improving the quality of frequency domain reconstruction; the calculation of the spatial domain consistency loss value ensures the structural similarity and detail expression of the reconstructed image in the spatial domain, helps to restore the local details and overall structure of the image, and makes the reconstructed image visually clearer and more accurate; the calculation of the reconstruction loss value measures the difference between the reconstructed image and the real image as a whole, ensures the quality of the reconstructed image both globally and locally, helps to balance the reconstruction effects in the frequency domain and spatial domain, and generates more natural and closer to the real high-resolution image; by calculating the loss values ​​in the frequency domain and spatial domain separately, the model can better utilize the information of the auxiliary modality to guide the reconstruction process. This multimodal fusion method helps to improve the model's processing capabilities for complex MRI data and generate more reliable reconstruction results. The clear loss value calculation provides a clear optimization direction for model training. By continuously adjusting the model parameters to minimize these loss values, the model can converge to the optimal solution faster, thereby improving training efficiency.

[0093] In some embodiments, the frequency domain consistency loss value, the spatial domain consistency loss value, and the reconstruction loss value are weightedly fused, including:

[0094] When the proportion of low-frequency components in the training low-resolution frequency domain data is greater than that of high-frequency components, the weight value of the frequency domain consistency loss value is greater than the weight value of the spatial domain consistency loss value;

[0095] When the proportion of low-frequency components in the training low-resolution frequency domain data is not greater than that of high-frequency components, the weight value of the spatial domain consistency loss value is greater than the weight value of the frequency domain consistency loss value;

[0096] Based on the weight value of the spatial domain consistency loss value, the weight value of the frequency domain consistency loss value and the weight value of the reconstruction loss value, the frequency domain consistency loss value, the spatial domain consistency loss value and the reconstruction loss value are weightedly fused.

[0097] The low-resolution training frequency domain data is analyzed to determine the proportion of low-frequency and high-frequency components. This analysis can be achieved through frequency domain energy distribution calculation or frequency domain feature statistics. When the low-frequency components in the training low-resolution frequency domain data are greater than the high-frequency components, the weight of the frequency domain consistency loss is increased so that it accounts for a larger proportion in the total loss function. This helps the model focus more on the accuracy of frequency domain reconstruction when low-frequency information dominates. Conversely, when the low-frequency components do not exceed the high-frequency components, the weight of the spatial domain consistency loss is increased so that it accounts for a larger proportion in the total loss function. This helps the model focus more on the recovery of spatial details and structures when high-frequency information is relatively abundant. Based on the adjusted spatial domain consistency loss weight, the frequency domain consistency loss weight, and the preset reconstruction loss weight, the frequency domain consistency loss, spatial domain consistency loss, and reconstruction loss are weighted and fused. The total loss function value is generated through weighted summation to guide the optimization of model parameters. The phased fusion of the attention mechanism is consistent with the logic of spectral energy transfer. In the early iterations, the structural attention mechanism (SP-former) prioritizes the fusion of the global low-frequency features of the auxiliary modality (such as T1); in the later iterations, it enhances the local high-frequency features of the target modality (such as T2).

[0098] This embodiment dynamically adjusts the loss weights so that the model can automatically balance the reconstruction priorities of the frequency domain and the spatial domain according to the characteristics of the input data. When low-frequency information dominates, the accuracy of frequency domain reconstruction is enhanced; when high-frequency information is abundant, the ability to recover spatial domain details is enhanced. This dynamic adjustment strategy is particularly suitable for multimodal MRI reconstruction tasks. It can fully utilize the complementary information between different modalities and improve the overall reconstruction performance. By combining the advantages of the frequency domain and the spatial domain, the model can generate more comprehensive and reliable reconstruction results.

[0099] The embodiments of the present application are described and illustrated below through preferred embodiments.

[0100] The preferred embodiment of the MRI image reconstruction method of the present application includes the following steps:

[0101] Step S301, data acquisition and preprocessing of multi-contrast MRI samples;

[0102] Step S302, constructing a low-frequency truncation-based degradation process;

[0103] Step S303: constructing a dual-domain collaborative recovery process;

[0104] Step S304, constructing dual-domain data consistency loss and reconstruction loss;

[0105] Step S305 : training and reasoning the MRI super-resolution framework based on the multi-contrast dual-domain cooperative diffusion model guided by low-frequency prior.

[0106] For step S301, the target modality (T2) in the multi-contrast MRI data and auxiliary modality (T1) Slice along the depth axis into 2D data and get , , perform standardization preprocessing;

[0107] in, These are three-dimensional (3D) T2-weighted high-resolution (HR) magnetic resonance images; 3D denotes three-dimensional data; T2 denotes the target modality (T2-weighted imaging); H, W, and D denote the height, width, and depth of the image. These are three-dimensional (3D) T1-weighted high-resolution (HR) magnetic resonance images; T1 denotes auxiliary modality (T1-weighted imaging). represents a two-dimensional (2D) T2-weighted image after slicing along the depth axis; represents a two-dimensional (2D) T1-weighted image after slicing along the depth axis, and d represents the feature dimension (scaling factor in the attention mechanism).

[0108] For step S302, the traditional diffusion model uses Gaussian noise as the degradation operator. Since low frequency is required as a priori, Figure 3 is a preferred flow chart of the MRI image reconstruction method according to the preferred embodiment of the present application, such as Figure 3 As shown: The target modality high resolution image (T2 HR) is Perform Fourier transform to obtain k-space data (i.e. initial frequency domain data). Among them, is the target modality high-resolution image, which is a high-resolution image belonging to the spatial domain (image domain); s represents the spatial domain (spatial domain); 0 represents the initial time step, and n represents the size of the image (such as n×n represents a square image). It means that the spatial domain image is converted into frequency domain (k-space) data through Fourier transform F, where k represents the frequency domain (k-space).

[0109] By defining a mask that gradually blocks high-frequency components from the edge to the center, the k-space truncation degradation is simulated to generate the k-space data at time step t. (Degraded k-space low-frequency data, i.e. low-resolution frequency domain data).

[0110] The degradation process of k-space is as follows: ;

[0111] In the above formula, represents low-resolution frequency domain data, represents element-wise multiplication, is the degradation mask of step t, which is used to cut off the high-frequency components of k-space. Its formula is as follows: ;

[0112] In the above formula, β t Represents the mask parameters (dynamically adjusted with the number of steps); (u, v) represents the pixel coordinates of the degradation mask in k-space, c represents the pixel center coordinates of the degradation mask, The side length of the k-space center region (low-frequency component) retained in the tth step is controlled and gradually decreases with the increase of the step number t to simulate the gradual loss of high-frequency information.

[0113] Again Perform inverse Fourier transform back to image domain , That is, the spatial domain low-resolution image, F represents Fourier transform, Represents the inverse transform. This process directly uses low-frequency information as the diffusion starting point to avoid invalid noise degradation. Specifically, the degradation process in the image domain is as follows:

[0114] ;

[0115] In the above formula, represents element-wise multiplication, For the The degradation mask of the step, represents the inverse Fourier transform, Represents a low-resolution image in the spatial domain.

[0116] For step S303, Figure 4 is a dual-domain cooperative diffusion model framework diagram of the MRI image reconstruction method according to the preferred embodiment of the present application, such as Figure 4 Shown: Constructing a dual-domain collaborative recovery process and introducing auxiliary modality high-resolution images (T1 HR) during the recovery process , for the k-space part, and After extracting features through the encoder, the structure prior attention mechanism (SP-former) is used to fuse features, and then restored to ; For the spatial domain There is no need to perform Fourier transform. After extracting features through the encoder, SP-former is used for feature fusion, and then the encoder is used to restore .

[0117] The dual-domain collaborative recovery process is as follows;

[0118] Building a recovery operator Reverse the degradation process D, and this operator can be parameterized by The neural network implementation is as follows:

[0119] ;

[0120] In the above formula, A restoration operator (neural network) with parameter θ, used to reverse the degradation process; x t Represents the input data of the current time step t. represents the reconstruction loss, which constrains the difference between the restored result and the true HR image.

[0121] Construct a structural attention mechanism (SP-former) to fuse the auxiliary modal features with the current time step features, see Figure 4 ,as follows:

[0122] ;

[0123] in, , , ;

[0124] In the above formula, , is a learnable attention weight (query, key, value), represents the downsampled features of the current time step, represents the downsampled features of the auxiliary modality (T1). Figure 4 In [1], MLP (Multi-Layer Perceptron) is a feedforward neural network structure in deep learning; Sigmoid function is a commonly used activation function in neural networks, with an S-shaped curve. Its core function is to map any real number to the interval (0, 1), and the output can be interpreted as a probability value.

[0125] For step S304, a dual-domain data consistency loss and a reconstruction loss are constructed, a spectral consistency loss is introduced to ensure cross-domain physical fidelity, a spatial domain consistency loss is combined to optimize visual fidelity, and a weighted fusion is performed with the reconstruction loss of the diffusion model; the details are as follows:

[0126] Based on the recovery operator The output of and physical imaging constraints define the dual-domain data consistency loss:

[0127] For the frequency domain:

[0128] ;

[0129] For the spatial domain:

[0130] ;

[0131] in, , ,The constrained dual domain maintains frequency domain consistency and visual consistency in the ,iterative restoration process, reducing the accumulated error.

[0132] In the above formula, Represents the frequency domain consistency loss, which constrains the reconstructed spectrum to be aligned with the true spectrum; Represents the spatial consistency loss, which constrains the reconstructed image to be aligned with the real image. , Represents the loss weight for adjusting frequency domain fidelity and spatial domain visual quality.

[0133] Construct the reconstruction loss:

[0134] For the frequency domain:

[0135] ;

[0136] For the spatial domain:

[0137] ;

[0138] The total loss function is defined as follows:

[0139] For the frequency domain:

[0140] ;

[0141] For the spatial domain:

[0142] ;

[0143] In the above formula, represents the total loss function in the frequency domain, combining reconstruction loss and consistency loss; Represents the total spatial loss function, which combines the reconstruction loss and the consistency loss. , is the weight adjustment coefficient, which is used to adjust the weight of the reconstruction loss and the dual-domain data consistency loss.

[0144] For step S305, the MRI super-resolution framework based on the multi-contrast dual-domain cooperative diffusion model guided by low-frequency prior is trained and inferred: for the training process, the standard T2 HR is degraded through step S302 and then restored through step S303, that is, the T2HR is generated into low-resolution data through the degradation process of step S302, and then the image is reconstructed through the dual-domain cooperative restoration process (combined with T1HR) in step S303, and the model parameters are optimized using the total loss function defined in step S304; for the inference process, T2 LR is directly used to restore T2 HR through step S303, that is, T2LR is directly input, and a high-resolution T2 image is generated through the dual-domain cooperative restoration process (combined with T1HR) in step S303 without the degradation step.

[0145] This embodiment also provides an MRI image reconstruction device for implementing the above-mentioned embodiments and preferred embodiments. Details already described will not be repeated. As used below, terms such as "module," "unit," and "subunit" may refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation using hardware, or a combination of software and hardware, is also possible and contemplated.

[0146] Figure 5 is a structural block diagram of an MRI image reconstruction device according to an embodiment of the present application, such as Figure 5 As shown, the device includes:

[0147] An image acquisition module 10 is used to acquire multi-contrast MRI sample data, including a target modality high-resolution image and an auxiliary modality high-resolution image;

[0148] The data processing module 20 is configured to perform frequency domain conversion on the target modality high-resolution image to obtain initial frequency domain data, truncate high-frequency components of the initial frequency domain data to generate low-resolution frequency domain data, and perform spatial domain conversion on the low-resolution frequency domain data to generate a spatial domain low-resolution image.

[0149] The image reconstruction module 30 is used to input the low-resolution frequency domain data and the low-resolution spatial domain image into the diffusion model, and use the diffusion model to extract features of the low-resolution frequency domain data and the low-resolution spatial domain image in the frequency domain and the spatial domain respectively to generate the current time step features; and use the structural attention mechanism to fuse the current time step features with the features of the auxiliary modality high-resolution image to obtain the reconstructed image of the target modality.

[0150] The above-mentioned data processing module 20 is also used to obtain a preset mask that blocks high-frequency components from the edge to the center according to a preset step size. Based on the initial frequency domain data, the high-frequency components are truncated by gradually reducing the mask of the frequency center area to generate low-resolution frequency domain data at different time steps.

[0151] The above-mentioned image reconstruction module 30 is also used for the current time step features including frequency domain time step features and spatial domain time step features;

[0152] In the frequency domain, encoder features are extracted from low-resolution frequency domain data to generate frequency domain time step features. The frequency domain time step features are fused with the frequency domain features of the auxiliary modality high-resolution image using the structural attention mechanism to generate high-frequency missing distribution features.

[0153] In the spatial domain, encoder features are extracted from the spatial domain low-resolution image to generate spatial domain time step features. The structural attention mechanism is used to fuse the spatial domain time step features with the spatial domain features of the auxiliary modality high-resolution image to generate local detail features.

[0154] Based on the high-frequency missing distribution characteristics and local detail features, a reconstructed image of the target modality is obtained.

[0155] The above-mentioned reconstructed image module 30 is also used for the structural attention mechanism to dynamically fuse the current time step features with the features of the auxiliary modality high-resolution image through the attention weight.

[0156] The MRI image reconstruction device further includes a training module; the training module is used to obtain training MRI sample data, including training target modality high-resolution images and training auxiliary modality high-resolution images;

[0157] Perform frequency domain conversion on the high-resolution image of the training target modality to obtain the initial frequency domain data for training, truncate the high-frequency components of the initial frequency domain data for training to generate low-resolution frequency domain data for training; perform spatial domain conversion on the low-resolution frequency domain data for training to generate low-resolution images in the spatial domain for training;

[0158] The training low-resolution frequency domain data and the training spatial domain low-resolution images are input into the initial diffusion model. The initial diffusion model is used to extract features of the training low-resolution frequency domain data and the training spatial domain low-resolution images in the frequency domain and spatial domain, respectively, to generate the features of the training current time step. The structural attention mechanism is then used to fuse the features of the training current time step with the features of the training auxiliary modality high-resolution images to reconstruct the training reconstructed image of the target modality.

[0159] The training reconstructed image includes frequency domain reconstruction data and spatial domain reconstruction data;

[0160] Constructing a dual-domain data consistency loss function and a reconstruction loss function; wherein the dual-domain data consistency loss function includes a frequency domain consistency loss function and a spatial domain consistency loss function;

[0161] Based on the dual-domain data consistency loss function and the reconstruction loss function, the frequency domain consistency loss value, the spatial domain consistency loss value and the reconstruction loss value are calculated;

[0162] The frequency domain consistency loss value, spatial domain consistency loss value and reconstruction loss value are weighted and fused to obtain the total loss function value; based on the total loss function value, the parameters of the initial diffusion model are iteratively optimized to generate a trained diffusion model.

[0163] The training module is further configured to calculate the difference between the frequency domain features of the reconstructed image and the frequency domain features of the corresponding training target modality high-resolution image based on the frequency domain consistency loss function, thereby generating a frequency domain consistency loss value.

[0164] Based on the spatial consistency loss function, the difference between the spatial features of the reconstructed image and the corresponding spatial features of the training target modality high-resolution image is calculated to generate a spatial consistency loss value;

[0165] Based on the reconstruction loss function, the difference in features between the training reconstructed image and the corresponding training target modality high-resolution image is calculated to generate a reconstruction loss value.

[0166] The above training module is also used for, when the proportion of low-frequency components in the training low-resolution frequency domain data is greater than that of high-frequency components, the weight value of the frequency domain consistency loss value is greater than the weight value of the spatial domain consistency loss value;

[0167] When the proportion of low-frequency components in the training low-resolution frequency domain data is not greater than that of high-frequency components, the weight value of the spatial domain consistency loss value is greater than the weight value of the frequency domain consistency loss value;

[0168] Based on the weight value of the spatial domain consistency loss value, the weight value of the frequency domain consistency loss value and the weight value of the reconstruction loss value, the frequency domain consistency loss value, the spatial domain consistency loss value and the reconstruction loss value are weightedly fused.

[0169] This embodiment further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0170] Optionally, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0171] Optionally, in this embodiment, the processor may be configured to execute the following steps through a computer program:

[0172] S1, acquire multi-contrast MRI sample data, including target modality high-resolution images and auxiliary modality high-resolution images;

[0173] S2, performing frequency domain conversion on the target modality high-resolution image to obtain initial frequency domain data, truncating the high-frequency components of the initial frequency domain data to generate low-resolution frequency domain data; performing spatial domain conversion on the low-resolution frequency domain data to generate a spatial domain low-resolution image;

[0174] S3, inputs the low-resolution frequency domain data and the low-resolution spatial domain image into the diffusion model, and uses the diffusion model to extract features of the low-resolution frequency domain data and the low-resolution spatial domain image in the frequency domain and spatial domain respectively to generate the current time step features; and uses the structural attention mechanism to fuse the current time step features with the features of the auxiliary modality high-resolution image to obtain the reconstructed image of the target modality.

[0175] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementation modes, and this embodiment will not be repeated here.

[0176] In addition, in conjunction with the MRI image reconstruction method in the above embodiments, the present application embodiment may provide a storage medium for implementation. The storage medium stores a computer program; when the computer program is executed by a processor, any one of the MRI image reconstruction methods in the above embodiments is implemented.

[0177] Those skilled in the art should understand that the various technical features of the above-described embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the various technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0178] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A MRI image reconstruction method, characterized in that: include: Acquire multi-contrast MRI sample data, including target modality high-resolution images and auxiliary modality high-resolution images; Performing frequency domain conversion on the target modality high-resolution image to obtain initial frequency domain data, and truncating high-frequency components of the initial frequency domain data to generate low-resolution frequency domain data; Performing spatial domain conversion on the low-resolution frequency domain data to generate a spatial domain low-resolution image; Inputting the low-resolution frequency domain data and the low-resolution spatial domain image into a diffusion model, and using the diffusion model to extract features of the low-resolution frequency domain data and the low-resolution spatial domain image in the frequency domain and the spatial domain, respectively, to generate current time step features; And using the structural attention mechanism, feature fusion is performed on the current time step features and the features of the auxiliary modality high-resolution image to obtain a reconstructed image of the target modality, including: The current time step feature includes a frequency domain time step feature and a spatial domain time step feature; In the frequency domain, encoder feature extraction is performed on the low-resolution frequency domain data to generate the frequency domain time step feature; the frequency domain time step feature is fused with the frequency domain feature of the auxiliary modality high-resolution image using the structural attention mechanism to generate a high-frequency missing distribution feature; In the spatial domain, encoder features are extracted from the spatial domain low-resolution image to generate the spatial domain time step features; the spatial domain time step features are fused with the spatial domain features of the auxiliary modality high-resolution image using the structural attention mechanism to generate local detail features; A reconstructed image of the target modality is acquired based on the high-frequency missing distribution feature and the local detail feature.

2. The MRI image reconstruction method according to claim 1, wherein: The step of truncating high-frequency components of the initial frequency domain data to generate low-resolution frequency domain data includes: Obtain a preset mask that blocks high-frequency components from the edge to the center according to a preset step size. Based on the initial frequency domain data, reduce the mask in the frequency center area according to the preset step size, cut off the high-frequency components, and generate the low-resolution frequency domain data at different time steps.

3. The MRI image reconstruction method according to claim 1, wherein: The structural attention mechanism dynamically fuses the features of the current time step with the features of the auxiliary modality high-resolution image through attention weights.

4. The MRI image reconstruction method according to claim 1, wherein: The method further comprises: Acquire training MRI sample data, including high-resolution images of the training target modality and high-resolution images of the training auxiliary modality; Performing frequency domain conversion on the training target modality high-resolution image to obtain training initial frequency domain data, truncating high-frequency components of the training initial frequency domain data to generate training low-resolution frequency domain data; performing spatial domain conversion on the training low-resolution frequency domain data to generate a training spatial domain low-resolution image; Inputting the training low-resolution frequency domain data and the training spatial domain low-resolution image into an initial diffusion model, using the initial diffusion model, performing feature extraction on the training low-resolution frequency domain data and the training spatial domain low-resolution image in the frequency domain and the spatial domain, respectively, to generate training current time step features; and using a structured attention mechanism, performing feature fusion on the training current time step features and the features of the training auxiliary modality high-resolution image to reconstruct a training reconstructed image of the target modality; The training reconstructed image includes frequency domain reconstructed data and spatial domain reconstructed data; Constructing a dual-domain data consistency loss function and a reconstruction loss function; wherein the dual-domain data consistency loss function includes a frequency domain consistency loss function and a spatial domain consistency loss function; Calculating a frequency domain consistency loss value, a spatial domain consistency loss value, and a reconstruction loss value based on the dual-domain data consistency loss function and the reconstruction loss function; The frequency domain consistency loss value, the spatial domain consistency loss value and the reconstruction loss value are weightedly fused to obtain a total loss function value; based on the total loss function value, the parameters of the initial diffusion model are iteratively optimized to generate a trained diffusion model.

5. The MRI image reconstruction method according to claim 4, characterized in that: The calculating of the frequency domain consistency loss value, the spatial domain consistency loss value, and the reconstruction loss value based on the dual-domain data consistency loss function and the reconstruction loss function includes: Based on the frequency domain consistency loss function, calculating the difference between the frequency domain reconstructed image features and the corresponding frequency domain features of the training target modality high-resolution image to generate a frequency domain consistency loss value; Based on the spatial consistency loss function, calculating the difference between the spatial reconstructed image features and the corresponding spatial features of the training target modality high-resolution image to generate a spatial consistency loss value; Based on the reconstruction loss function, the difference between the features of the training reconstructed image and the corresponding high-resolution image of the training target modality is calculated to generate a reconstruction loss value.

6. The MRI image reconstruction method according to claim 4, wherein: The weighted fusion of the frequency domain consistency loss value, the spatial domain consistency loss value and the reconstruction loss value includes: When the proportion of low-frequency components in the training low-resolution frequency domain data is greater than that of high-frequency components, the weight value of the frequency domain consistency loss value is greater than the weight value of the spatial domain consistency loss value; When the proportion of low-frequency components in the training low-resolution frequency domain data is not greater than that of high-frequency components, the weight value of the spatial domain consistency loss value is greater than the weight value of the frequency domain consistency loss value; Based on the weight value of the spatial domain consistency loss value, the weight value of the frequency domain consistency loss value and the weight value of the reconstruction loss value, the frequency domain consistency loss value, the spatial domain consistency loss value and the reconstruction loss value are weightedly fused.

7. An MRI image reconstruction device, characterized in that: include: An image acquisition module, used to acquire multi-contrast MRI sample data, including target modality high-resolution images and auxiliary modality high-resolution images; a data processing module, configured to perform frequency domain conversion on the target modality high-resolution image to obtain initial frequency domain data, and to truncate high-frequency components of the initial frequency domain data to generate low-resolution frequency domain data; Performing spatial domain conversion on the low-resolution frequency domain data to generate a spatial domain low-resolution image; an image reconstruction module, configured to input the low-resolution frequency domain data and the spatial domain low-resolution image into a diffusion model, and utilize the diffusion model to extract features of the low-resolution frequency domain data and the spatial domain low-resolution image in the frequency domain and the spatial domain, respectively, to generate features of the current time step; And using the structural attention mechanism, feature fusion is performed on the current time step features and the features of the auxiliary modality high-resolution image to obtain a reconstructed image of the target modality, including: The current time step feature includes a frequency domain time step feature and a spatial domain time step feature; In the frequency domain, encoder feature extraction is performed on the low-resolution frequency domain data to generate the frequency domain time step feature; the frequency domain time step feature is fused with the frequency domain feature of the auxiliary modality high-resolution image using the structural attention mechanism to generate a high-frequency missing distribution feature; In the spatial domain, encoder features are extracted from the spatial domain low-resolution image to generate the spatial domain time step features; the spatial domain time step features are fused with the spatial domain features of the auxiliary modality high-resolution image using the structural attention mechanism to generate local detail features; A reconstructed image of the target modality is acquired based on the high-frequency missing distribution feature and the local detail feature.

8. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to perform the MRI image reconstruction method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute the MRI image reconstruction method according to any one of claims 1 to 6 when executed.

Images