A method and apparatus for super-resolution reconstruction in magnetic resonance imaging

Abstract

This application relates to a method and apparatus for super-resolution reconstruction of magnetic resonance imaging (MRI), applied in the field of medical image processing technology. The medical image processing method includes: acquiring at least one low-resolution target contrast MRI image and at least one high-resolution reference contrast MRI image corresponding to the super-resolution MRI image to be reconstructed, and converting them respectively into a preset Fourier space to obtain corresponding Fourier space data; obtaining Fourier space sequence data according to a preset sorting method, and inputting it into a target deep learning model to obtain predicted high-frequency region data; the target deep learning model adopts an encoding-decoding structure; and generating a super-resolution reconstructed target contrast MRI image based on the predicted high-frequency region data and the low-resolution target space data. This application improves the accuracy and robustness of generating high-resolution images in MRI super-resolution methods.

Description

Technical Field

[0001] This application relates to the field of medical image processing technology, and in particular to a method and apparatus for super-resolution reconstruction of magnetic resonance imaging. Background Technology

[0002] Magnetic Resonance Imaging (MRI) is an important medical imaging technique that provides rich contrast information of soft tissues without ionizing radiation. However, MRI imaging is relatively slow, and acquiring high-resolution (HR) images typically requires a long scan time, which can lead to motion artifacts, patient discomfort, and limited clinical throughput. To shorten scan time, low-resolution (LR) images are often acquired, but this sacrifices image detail and affects diagnostic accuracy.

[0003] MRI super-resolution (SR) technology aims to reconstruct high-resolution (HR) images from low-resolution (LR) images, and is a key approach to resolving the aforementioned contradictions. Existing MRI SR methods include interpolation-based methods, reconstruction-based methods (such as compressed sensing), and learning-based methods (especially deep learning). However, these methods still cannot fully utilize the characteristics of MRI signals, resulting in the need to improve the accuracy and robustness of generating high-resolution images.

[0004] There is currently no effective solution to the problem that the accuracy and robustness of generating high-resolution images in existing MRI super-resolution methods still need to be improved. Summary of the Invention

[0005] This embodiment provides a magnetic resonance imaging super-resolution reconstruction method and apparatus to address the issue that the accuracy and robustness of generating high-resolution images in related MRI super-resolution methods still need to be improved.

[0006] Firstly, this embodiment provides a magnetic resonance imaging super-resolution reconstruction method, the method comprising:

[0007] Acquire at least one low-resolution target contrast magnetic resonance image corresponding to the super-resolution magnetic resonance imaging to be reconstructed, and at least one high-resolution reference contrast magnetic resonance image.

[0008] The low-resolution target contrast magnetic resonance image and the high-resolution reference contrast magnetic resonance image are respectively converted from the original image space to a preset Fourier space to obtain the corresponding Fourier space data; the Fourier space data includes high-resolution reference space data and low-resolution target space data.

[0009] The Fourier space data is sorted according to a preset sorting method to obtain Fourier space sequence data.

[0010] The Fourier space sequence data is input into the target deep learning model to obtain the predicted high-frequency region data; the target deep learning model is trained based on a preset deep learning model with an encoding and decoding structure, and is used to generate high-frequency region data of target contrast in the preset Fourier space.

[0011] Based on the predicted high-frequency region data and the low-resolution target spatial data, a super-resolution reconstructed target contrast magnetic resonance image is generated.

[0012] In some embodiments, sorting the Fourier space data according to a preset sorting method to obtain Fourier space sequence data includes:

[0013] Acquire the high-resolution reference space data and the low-resolution target space data, and the corresponding multiple spatial points; calculate the multiple spatial distances of the multiple spatial points from the center point in the preset Fourier space;

[0014] Based on the multiple spatial distances, the multiple spatial points are arranged in a vector sequence, and based on a preset embedding strategy, one-dimensional sequence data in Fourier space is determined.

[0015] In some embodiments, sorting the Fourier space data according to a preset sorting method to obtain Fourier space sequence data includes:

[0016] According to the preset blocks, the high-resolution reference spatial data and the low-resolution target spatial data are divided into multiple spatial data blocks;

[0017] Calculate the distance between the center point of the plurality of spatial data blocks and the center point of the plurality of spatial blocks in the preset Fourier space;

[0018] Based on the distances between the multiple spatial blocks, the multiple spatial data blocks are arranged in a vector sequence, and based on a preset embedding strategy, the Fourier space multidimensional sequence data is determined.

[0019] In some embodiments, the method further includes:

[0020] The low-resolution target spatial data is determined according to a preset sampling strategy.

[0021] In some of these embodiments, the preset deep learning model includes an encoder and a decoder;

[0022] The step of inputting the Fourier spatial sequence data into the target deep learning model to obtain predicted high-frequency region data includes:

[0023] The high-resolution reference space data, processed according to the preset sorting method, is input into the encoder to generate a context representation;

[0024] The context representation and the low-resolution target spatial data processed according to a preset sorting method are input into the decoder to generate predicted high-frequency region data.

[0025] In some embodiments, generating a super-resolution reconstructed target contrast magnetic resonance image based on the predicted high-frequency region data and the low-resolution target spatial data includes:

[0026] The high-frequency region data and the low-resolution target space data are merged to obtain the predicted high-resolution target space data;

[0027] The predicted high-resolution target spatial data is transformed from a preset Fourier space to the original image space to generate a super-resolution reconstructed target contrast magnetic resonance image.

[0028] In some embodiments, the method further includes:

[0029] Extract high-frequency region data of targets from real high-resolution target training images in a preset training set; the preset training set also includes low-resolution target training images and high-resolution reference training images.

[0030] The low-resolution target training image and the high-resolution reference training image are input into a preset deep learning model to determine the high-frequency region data for predicting the target contrast in Fourier space.

[0031] Based on a preset loss function, the loss values ​​of the target high-frequency region data and the predicted high-frequency region data are determined;

[0032] Based on the preset gradient descent algorithm and the loss value, the preset deep learning model is trained to obtain the target deep learning model.

[0033] Secondly, this embodiment provides a magnetic resonance imaging super-resolution reconstruction device, which includes: an image acquisition module, a spatial transformation module, a sorting module, and a reconstruction module;

[0034] The image acquisition module is used to acquire at least one low-resolution target contrast magnetic resonance image corresponding to the super-resolution magnetic resonance imaging to be reconstructed, and at least one high-resolution reference contrast magnetic resonance image.

[0035] The spatial transformation module is used to transform the low-resolution target contrast magnetic resonance image and the high-resolution reference contrast magnetic resonance image from the original image space to a preset Fourier space, respectively, to obtain corresponding Fourier space data; the Fourier space data includes high-resolution reference space data and low-resolution target space data;

[0036] The sorting module is used to sort the Fourier space data according to a preset sorting method to obtain Fourier space sequence data.

[0037] The reconstruction module is used to input the Fourier space sequence data into the target deep learning model to obtain predicted high-frequency region data; the target deep learning model is trained based on a preset deep learning model with an encoding and decoding structure, and is used to generate high-frequency region data of target contrast in a preset Fourier space; it is also used to generate a super-resolution reconstructed target contrast magnetic resonance image based on the predicted high-frequency region data and the low-resolution target space data.

[0038] Thirdly, this embodiment provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the magnetic resonance imaging super-resolution reconstruction method described in the first aspect above.

[0039] Fourthly, this embodiment provides a storage medium storing a computer program that, when executed by a processor, implements the magnetic resonance imaging super-resolution reconstruction method described in the first aspect above.

[0040] Compared with related technologies, the magnetic resonance imaging super-resolution reconstruction method and apparatus provided in this embodiment transforms a low-resolution target contrast image and a high-resolution reference contrast image from the original image space to Fourier space (K-space). Then, in Fourier space, a deep learning model with an encoding-decoding structure is used to predict the high-frequency region data of the target contrast. After prediction, the predicted data is transformed back from Fourier space to the original image space. By utilizing the frequency domain characteristics of K-space and a method that fuses information from the target contrast image and the reference contrast image, the generation of high-frequency information of the target contrast is guided, ensuring that the generated high-frequency information is consistent with the low-frequency information, thus improving the accuracy and robustness of reconstructing high-resolution images.

[0041] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0042] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0043] Figure 1 This is a hardware structure block diagram of the terminal of the magnetic resonance imaging super-resolution reconstruction method provided in the embodiments of this application;

[0044] Figure 2 This is a flowchart of the magnetic resonance imaging super-resolution reconstruction method provided in the embodiments of this application;

[0045] Figure 3 This is a schematic diagram of the process of obtaining a vector representation by adding various embeddings provided in the embodiments of this application;

[0046] Figure 4 This is a schematic diagram of the spatial high and low frequency region division provided in an embodiment of this application;

[0047] Figure 5 This is a schematic diagram of a magnetic resonance imaging super-resolution computer system provided in this specific embodiment;

[0048] Figure 6 This is a flowchart of the multi-contrast magnetic resonance imaging super-resolution method based on K-space Transformer provided in this specific embodiment;

[0049] Figure 7 This is a flowchart of obtaining the high-frequency Fourier spatial sequence of the predicted target provided in this specific embodiment. Detailed Implementation

[0050] To better understand the purpose, technical solution, and advantages of this application, the application is described and illustrated below in conjunction with the accompanying drawings and embodiments.

[0051] Unless otherwise defined, the technical or scientific terms used in this application shall have the general meaning understood by one of ordinary skill in the art to which this application pertains. Words such as “a,” “an,” “an,” “the,” “the,” and “these” used in this application do not indicate quantitative limitation and may be singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps or modules (units) is not limited to the listed steps or modules (units) but may include steps or modules (units) not listed, or may include other steps or modules (units) inherent to these processes, methods, products, or devices. Words such as “connected,” “linked,” and “coupled” used in this application are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. Normally, the character " / " indicates that the objects before and after it are in an "or" relationship. The terms "first," "second," "third," etc., used in this application are merely to distinguish similar objects and do not represent a specific order of objects.

[0052] The method embodiments provided in this example can be executed on a terminal, computer, or similar computing device. For example, it can run on a terminal. Figure 1 This is a hardware structure block diagram of the terminal for the magnetic resonance imaging super-resolution reconstruction method provided in this application embodiment. For example... Figure 1 As shown, a terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 and a memory 104 for storing data are also included. The processor 102 may be, but is not limited to, a microprocessor (MCU) or a programmable logic device (FPGA). The terminal may also include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that… Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the terminal described above. For example, the terminal may also include components that are larger than... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown are illustrated.

[0053] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the magnetic resonance imaging super-resolution reconstruction method in this embodiment. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, thereby implementing the above-described 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 instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0054] The transmission device 106 is used to receive or send data via a network. This network includes a wireless network provided by the terminal's communication provider. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 can be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0055] Magnetic Resonance Imaging (MRI) is a medical diagnostic technique that uses the principle of nuclear magnetic resonance to image the inside of the human body. It can provide detailed images of internal structures, especially with excellent contrast in soft tissues, and therefore has wide applications in the diagnosis of diseases of the nervous system, musculoskeletal system, cardiovascular system, and other systems.

[0056] Super-resolution (SR) refers to a series of techniques used to improve the resolution of images or videos. For MRI images, super-resolution techniques can be used to enhance image sharpness and detail, thereby helping doctors make more accurate diagnoses. These techniques can be implemented through different methods, including but not limited to: interpolation-based methods, reconstruction-based methods, and deep learning-based methods.

[0057] Current deep learning-based methods, especially convolutional neural networks (CNNs), have achieved significant success in SR tasks. However, most existing methods process data in the image domain, potentially failing to fully utilize the inherent characteristics of MRI signals in the Fourier-K space (frequency domain). MRI K-space data contains frequency information of the image, where each point represents information at a specific frequency in the image, with the central region representing low-frequency information (general image outline and contrast) and the peripheral region representing high-frequency information (image details and edges).

[0058] Furthermore, clinical MRI examinations typically acquire multiple contrast images of the same location (e.g., T1-weighted, T2-weighted, FLAIR, proton density-weighted, etc.). These images with different contrasts possess inherent anatomical consistency and information complementarity. Existing multi-contrast SR methods attempt to utilize this complementarity, but often employ simple channel stitching or shallow fusion strategies in the image domain. This makes it difficult to effectively and deeply mine and utilize the complex correlations between images of different contrasts in K-space to guide the generation of high-frequency information from the target contrast image.

[0059] To address the problems of insufficient utilization of K-space information, inefficient fusion of multi-contrast information, and difficulty in generating high-frequency details in existing magnetic resonance imaging (MRI) methods, there is an urgent need for an MRI super-resolution method that can operate directly in K-space and efficiently fuse multi-contrast information to accurately generate high-frequency details of the target contrast.

[0060] This embodiment provides a magnetic resonance imaging super-resolution reconstruction method. Figure 2 This is a flowchart of the magnetic resonance imaging super-resolution reconstruction method provided in the embodiments of this application, such as... Figure 2 As shown, the process includes the following steps:

[0061] Step S210: Obtain at least one low-resolution target contrast magnetic resonance image and at least one high-resolution reference contrast magnetic resonance image corresponding to the super-resolution magnetic resonance imaging to be reconstructed.

[0062] When it is necessary to improve high-frequency detail processing of the super-resolution magnetic resonance imaging to be reconstructed, it is first necessary to acquire at least one low-resolution target contrast magnetic resonance image and at least one registered high-resolution reference contrast magnetic resonance image.

[0063] Specifically, target contrast refers to the type of magnetic resonance imaging (MRI) image that needs to be reconstructed at super-resolution. Reference contrast refers to spatially registered high-resolution MRI images with different imaging contrasts, used as supplementary information to guide the reconstruction of the target image. For example, a low-resolution target contrast MRI image is set as a low-resolution T2-weighted image; a high-resolution reference contrast MRI image is set as a high-resolution T1-weighted image and a high-resolution fluid attenuation inversion recovery image. The method of setting different contrast images can be determined according to the actual MRI scenario and is not specifically limited here.

[0064] Furthermore, T1-weighted imaging generates images based on the longitudinal relaxation time (T1) of tissues. By using shorter repetition times (TR) and echo times (TE), this technique can highlight T1 differences between different tissues. Adipose tissue appears as high signal intensity (bright) on T1-weighted images, while water or cerebrospinal fluid appears as low signal intensity (dark). T1WI is particularly useful for displaying anatomical structures.

[0065] T2-weighted imaging relies on transverse relaxation time (T2). It employs longer TR and TE to enhance the contrast of T2 values ​​between tissues. In contrast to T1-weighted imaging, water or cerebrospinal fluid appears as a high signal intensity (brightness) on T2-weighted images, while fat and other soft tissues show varying degrees of brightness depending on their water content. T2-weighted imaging is helpful in detecting pathological changes such as edema and inflammation.

[0066] Fluid attenuated inversion recovery imaging (FLAIR) is a special T2-weighted imaging technique that uses a selective inversion pulse to suppress the signal of free water, thereby reducing the influence of fluids such as cerebrospinal fluid on the image. This technique effectively eliminates high signal intensity around the ventricular system, making lesions near these areas more apparent, and is particularly useful for identifying white matter diseases and subcortical lesions.

[0067] Step S220: The low-resolution target contrast magnetic resonance image and the high-resolution reference contrast magnetic resonance image are converted from the original image space to a preset Fourier space to obtain the corresponding Fourier space data; the Fourier space data includes high-resolution reference space data and low-resolution target space data.

[0068] Among them, since at least one low-resolution target contrast magnetic resonance image and at least one high-resolution reference contrast magnetic resonance image are located in the original image space, and since they are in the image domain, the structural characteristics of frequency information cannot be determined, and high-frequency details cannot be generated.

[0069] Therefore, in this embodiment, at least one low-resolution target contrast magnetic resonance image and at least one high-resolution reference contrast magnetic resonance image located in the original image space (i.e., the image domain) are transformed from the original image space to a preset Fourier space to obtain corresponding Fourier space data. The Fourier space data includes high-resolution reference space data corresponding to the high-resolution reference contrast magnetic resonance image and low-resolution target space data corresponding to the low-resolution target contrast magnetic resonance image.

[0070] The method for transforming the original image space into the preset Fourier space is specifically set as Fourier transform. In practical applications, Fourier transform and its variants can both achieve the effect of transforming to the frequency domain. Here, no specific Fourier transform method is limited.

[0071] Step S230: Sort the Fourier space data according to the preset sorting method to obtain Fourier space sequence data.

[0072] In this process, after obtaining Fourier space data with frequency domain characteristics, the Fourier space data is sorted within the Fourier space to achieve preprocessing and serialization, thereby obtaining Fourier space sequence data. This Fourier space sequence data is a sequence suitable for input into a deep learning model for processing. For example, the preset sorting method includes: expanding and sorting the spatial points corresponding to the Fourier space data in one dimension based on frequency; and also includes performing two-dimensional / three-dimensional serialization processing based on the patches corresponding to the Fourier space data. No specific limitations are imposed on the particular sorting method here.

[0073] Step S240: Input the Fourier space sequence data into the target deep learning model to obtain the predicted high-frequency region data; the target deep learning model is trained based on a preset deep learning model with an encoding and decoding structure, and is used to generate the high-frequency region data of the target contrast in the preset Fourier space.

[0074] The process involves inputting preprocessed and serialized Fourier space data into a trained target deep learning model, and then generating predicted high-frequency region data through the encoding and decoding structure in the target deep learning model.

[0075] Step S250: Based on the predicted high-frequency region data and the low-resolution target spatial data, generate a super-resolution reconstructed target contrast magnetic resonance image.

[0076] In this process, after predicting the high-frequency region data, the predicted high-frequency region data and the low-resolution target spatial data are integrated according to the sequence to obtain the reconstructed target contrast sequence data. Subsequently, the reconstructed target contrast sequence data is transformed from Fourier space to the original image space to finally obtain the reconstructed high-resolution target contrast magnetic resonance image, i.e., the super-resolution reconstructed target contrast magnetic resonance image.

[0077] Through the above steps, the low-resolution target contrast image and the high-resolution reference contrast image are transformed from the original image space to Fourier space. Then, in Fourier space, a deep learning model with an encoding-decoding structure is used to predict the high-frequency region data of the target contrast. After prediction, the predicted data is transformed back from Fourier space to the original image space. By utilizing the frequency domain characteristics of Fourier space and combining the information from the target contrast image and the reference contrast image, the generation of high-frequency information guiding the target contrast is improved, ensuring consistency between the generated high-frequency information and the low-frequency information, thus enhancing the accuracy and robustness of the reconstructed high-resolution image.

[0078] In some embodiments, Fourier space data is sorted according to a preset sorting method to obtain Fourier space sequence data, including: acquiring high-resolution reference space data and low-resolution target space data, and corresponding multiple spatial points; calculating multiple spatial distances between the multiple spatial points and the center point in the preset Fourier space; arranging the multiple spatial points in a vector sequence according to the multiple spatial distances, and determining one-dimensional Fourier space sequence data based on a preset embedding strategy.

[0079] In this embodiment, the Fourier space data is expanded in one dimension based on frequency to obtain a one-dimensional sequence of Fourier space data, where frequency refers to the distance from the center of the Fourier space. Specifically, multiple spatial points corresponding to the high-resolution reference spatial data and the low-resolution target spatial data in the Fourier space data are determined, and multiple spatial distances from the preset center point of the Fourier space are calculated. Then, the spatial points are sorted in ascending order or in lexicographical order according to their coordinates.

[0080] For example, the spatial distance r from each spatial point (kx, ky) or (kx, ky, kz) to the K-space center (0,0) or (0,0,0) can be calculated. All spatial points are then sorted in ascending order based on the Euclidean distance r. To ensure the uniqueness of the sorting results, for spatial points with the same distance, such as those located on the same circle or sphere, a secondary sorting rule can be adopted, such as sorting in ascending order by polar coordinate angle (atan2(ky, kx)) or by lexicographical order of coordinates (kx first, ky second, kz third), finally obtaining a one-dimensional sequence of Fourier space data, that is, a one-dimensional K-space point sequence k[0], k[1], ..., where the first part of the data corresponds to low-frequency information and the latter part corresponds to high-frequency information.

[0081] In some embodiments, the Fourier space data is sorted according to a preset sorting method to obtain Fourier space sequence data. The method further includes: dividing the high-resolution reference space data and the low-resolution target space data into multiple spatial data blocks according to preset blocks; calculating the distance between the center point of the multiple spatial data blocks and the center point of the preset Fourier space in multiple spatial blocks; arranging the multiple spatial data blocks in a vector sequence according to the distance between the multiple spatial blocks; and determining the Fourier space multidimensional sequence data based on a preset embedding strategy.

[0082] In addition to the frequency-based one-dimensional sequence expansion method described in the above embodiments to obtain one-dimensional sequence data in Fourier space, two-dimensional or three-dimensional sequence data in Fourier space can also be obtained based on patches. Specifically, Fourier space can be regarded as an "image" and divided into small patches for processing to capture local frequency spatial correlations and reduce sequence length.

[0083] First, the size of the blocks is predefined, such as 2D PxP (e.g., 4x4) or 3D PxPxP. Based on the predefined blocks, the high-resolution reference spatial data and low-resolution target spatial data are divided into multiple spatial data blocks, which are typically non-overlapping (though some overlap is possible). Then, the order of these spatial data blocks in the sequence needs to be determined. The distance from the center of each block to the center of the Fourier space is calculated, and the blocks are sorted according to this distance (the uniqueness of the sorting of blocks with the same distance also needs to be handled). Finally, a block sequence is obtained: Block[0], Block[1], ... The spatial data points within each block are subsequently processed (e.g., flattened) and embedded to obtain a multidimensional sequence of Fourier space data.

[0084] In determining the sequence data in Fourier space (K-space), the sorted Fourier space data points or spatial data patches are represented as vector sequences (Tokens). For one-dimensional sequence data in Fourier space obtained by one-dimensional expansion, each Token typically represents a Fourier space point (containing its complex value information); for patch-based methods, each Token represents a patch (containing the complex value information of all points within the patch, usually first flattened into a vector). Each Token not only contains spatial data information (complex values ​​or flattened vectors) but also needs to incorporate its spatial location / frequency information.

[0085] In some of these embodiments, complex values ​​(real and imaginary parts) are concatenated with encoded positional information, such as position embedding, or a token vector is obtained through linear layer mapping.

[0086] Specifically, Figure 3 This is a schematic diagram illustrating the process of obtaining a vector representation by adding various embeddings provided in the embodiments of this application. (Reference) Figure 3 Data embedding: The data portion of each K-space token (e.g., complex values ​​or flattened vectors of patches), i.e. Fourier space sequence data, is mapped to a high-dimensional vector space through a linear layer to obtain a data vector.

[0087] Positional Embedding: Since the position (frequency) of a point in K-space or the position of a patch is crucial, positional embedding is needed to encode this spatial / frequency information. For one-dimensional sequence data in a one-dimensional expanded Fourier space, learnable positional embeddings based on the patch's index in the one-dimensional sequence can be used, or positional encodings calculated based on the patch's original K-space coordinates (kx, ky, kz) can be used (such as Fourier feature embeddings, which map the coordinates to a high-dimensional sine / cosine feature space, effectively representing the continuity of the frequency domain). For two-dimensional / three-dimensional patch sequences, positional embeddings that represent the patch's two-dimensional / three-dimensional position in the original K-space grid need to be designed, such as learnable 2D / 3D positional embeddings or Fourier feature embeddings based on the patch's center coordinates, ultimately yielding a position vector.

[0088] Modality Embedding: To distinguish low-frequency information between different reference contrasts and target contrasts, modality embedding can be introduced to ultimately obtain modality vectors. Specifically, a specific embedding vector can be learned for each input modality (such as T1, T2_LF, FLAIR, etc.) and added to the token embedding of the corresponding modality.

[0089] The data vectors, position vectors, and optional modal vectors obtained from data embedding, position embedding, and modal embedding are added or concatenated and mapped to form the input deep learning model. For example, the final token sequence is input into the Transformer model.

[0090] In some embodiments, the method further includes: determining low-resolution target spatial data according to a preset sampling strategy. Here, low-resolution target spatial data can be understood as low-frequency region data within high-resolution target spatial data.

[0091] in, Figure 4 This is a schematic diagram of the spatial high and low frequency region division provided in an embodiment of this application, for reference. Figure 4 Low-resolution target spatial data, i.e. low-frequency region data in high-resolution target spatial data, is usually the data in the central part of K-space. The size of its data region is determined by the sampling range of the low-resolution target contrast magnetic resonance image. The sampling range is determined according to the actual imaging sampling and is not specifically limited here.

[0092] In some of these embodiments, generating a super-resolution reconstructed target contrast magnetic resonance image based on predicted high-frequency region data and low-resolution target spatial data includes: merging the high-frequency region data and low-resolution target spatial data to obtain predicted high-resolution target spatial data; and converting the predicted high-resolution target spatial data from a preset Fourier space to the original image space to generate a super-resolution reconstructed target contrast magnetic resonance image.

[0093] In this process, after the high-frequency region data is predicted by the target deep learning model, the high-frequency region data and the low-resolution target space data are merged to form complete predicted high-resolution target space data.

[0094] Additionally, low-resolution target spatial data is used to replace any output that might be generated in the low-frequency region in the predicted high-frequency region data, in order to ensure the accuracy of the low-frequency information.

[0095] Subsequently, an inverse Fourier transform, such as a two-dimensional or three-dimensional fast inverse Fourier transform, is performed on the predicted high-resolution target spatial data to transform the predicted high-resolution target spatial data from Fourier space to the original image space. That is, the frequency domain data is converted into the image domain, ultimately obtaining the super-resolution reconstructed target contrast magnetic resonance image.

[0096] In some embodiments, the preset deep learning model includes an encoder and a decoder; inputting Fourier space sequence data into the target deep learning model to obtain predicted high-frequency region data includes: inputting high-resolution reference space data processed according to a preset sorting method into the encoder to generate a context representation; inputting the context representation and low-resolution target space data processed according to a preset sorting method into the decoder to generate predicted high-frequency region data.

[0097] The preset deep learning model is based on an encoder-decoder structure, including an encoder and a decoder. Specifically, it adopts an Encoder-Decoder Transformer model. This model can be based on the standard Transformer architecture, or it can be adaptively modified for the complex numerical characteristics of K-space data, such as using complex number operations or specific activation functions.

[0098] The encoder takes as input a preprocessed and embedded sequence of reference contrast K-space data, which corresponds to the sequence of high-resolution reference space data. Through a multi-layer self-attention mechanism and a feedforward network, it learns the complex dependencies and information representations within the high-resolution reference space data and between different reference contrasts, generating contextual representations rich in complementary information.

[0099] The input to the decoder is the context representation of the encoder output, and the known low-frequency K-space data sequence of the target contrast, i.e. the sequence corresponding to the low-resolution target spatial data, as the "context" or "prompt".

[0100] The decoder can employ an autoregressive approach, generating a sequence of HF tokens corresponding to one high-frequency region at a time, relying on previously generated tokens; or a non-autoregressive approach, generating sequence of HF tokens corresponding to all high-frequency region data all at once or in batches. This is generally faster but may sacrifice some accuracy. Regardless of the approach, the generation process must maintain consistency with the predefined serialization order of the high-frequency regions in the K-space. For example, if sorted by frequency, tokens are generated from low to high frequency; if sorted by patch, tokens are generated in the order of the patch sequence.

[0101] The decoder includes a self-attention layer and a cross-attention layer. The self-attention layer is used to process the sequence corresponding to the low-resolution target spatial data and the sequence corresponding to the predicted high-frequency region data generated by the decoder, as well as the internal dependency between the two.

[0102] The Cross-Attention Layer is used to interact with the current state of the decoder (based on the sequence corresponding to the low-resolution target spatial data and the sequence corresponding to the predicted high-frequency region data generated by the decoder) and the reference contrast context representation output by the encoder, "translating" or extracting information from the reference contrast information that helps generate the target high-frequency region data.

[0103] The decoder output is the sequence corresponding to the predicted high-resolution target spatial data, that is, the token sequence corresponding to the target contrast high-frequency K-space data k_tar_HF_pred.

[0104] In some embodiments, the method for training a preset deep learning model to obtain a target deep learning model includes: extracting target high-frequency region data from real high-resolution target training images in a preset training set; the preset training set also includes low-resolution target training images and high-resolution reference training images; inputting the low-resolution target training images and high-resolution reference training images into the preset deep learning model to determine the predicted high-frequency region data of target contrast in Fourier space; determining the loss value of the target high-frequency region data and the predicted high-frequency region data based on a preset loss function; and training the preset deep learning model based on the loss value using a preset gradient descent algorithm to obtain the target deep learning model.

[0105] Supervised learning is performed using paired low-resolution target training images (LR target images), high-resolution reference training images (HR reference images), and ground truth target training images (ground truth, I_tar_HR_gt). The ground truth target training image I_tar_HR_gt is transformed into K-space to obtain k_tar_HR_gt, and the high-frequency region data k_tar_HF_gt of its high-frequency component is extracted, corresponding to the high-frequency region to be predicted.

[0106] The loss function is used to define the loss function to measure the difference between the predicted high-frequency region data of the target contrast, i.e., the predicted high-frequency K-space data k_tar_HF_pred and the target high-frequency region data of the real high-resolution target training image, i.e., the real high-frequency data k_tar_HF_gt.

[0107] For example, the loss function includes: K-space loss: preferably L1 loss, which is more robust to outliers than L2 loss, calculating the difference between predicted high frequencies and true high frequencies: Loss_k = ||k_tar_HF_pred - k_tar_HF_gt||_1. This loss only applies to the high-frequency regions that require model generation.

[0108] Image domain loss: L1 or L2 loss, calculating the difference between the reconstructed image I_tar_HR_pred and the real high-resolution target training image I_tar_HR_gt: Loss_img=||I_tar_HR_pred-I_tar_HR_gt||_1 or Loss_img=||I_tar_HR_pred-I_tar_HR_gt||_2^2, where Loss_img represents the image domain loss.

[0109] Combined loss: The K-space loss and the image domain loss are usually combined in a weighted manner: Loss_total = lambda_k × Loss_k + lambda_img × Loss_img, where Loss_k represents the K-space loss, Loss_img represents the image domain loss, and lambda_k and lambda_img represent weight factors.

[0110] Furthermore, adversarial loss (GAN Loss) or perceptual loss can be selectively added to enhance the visual realism and detail of the image.

[0111] The parameters of the Transformer model are trained by minimizing the total loss function using an optimizer employing gradient descent optimization algorithms (such as Adam, AdamW).

[0112] The magnetic resonance imaging super-resolution reconstruction method provided in this application operates directly in K-space, which is more in line with the physical process of MRI imaging and can better utilize the frequency structure characteristics of K-space to generate high-frequency details.

[0113] The present embodiment will be described and explained below through specific examples.

[0114] This specific embodiment provides a multi-contrast magnetic resonance imaging super-resolution method based on K-space Transformer. This method can be implemented in software and run on a computer system equipped with sufficient computing resources (such as GPU, which needs to support parallel computing frameworks such as CUDA). Figure 5 This is a schematic diagram of a magnetic resonance imaging super-resolution computer system provided in this specific embodiment. (Reference) Figure 5 The system may include the following functional modules: data interface and preprocessing module, Fourier space processing module, Fourier space model module, reconstruction and output module, and model management and training module.

[0115] The multi-contrast magnetic resonance images are input into the data interface and processing module. These multi-contrast magnetic resonance images include at least one low-resolution target contrast magnetic resonance image and at least one high-resolution reference contrast magnetic resonance image, as described in the preceding embodiments.

[0116] Data interface and preprocessing module: Responsible for receiving multi-contrast magnetic resonance image data in formats such as DICOM or NIfTI, performing image registration when needed, and performing preprocessing such as intensity normalization, and then passing the processed data to the subsequent Fourier space processing module.

[0117] The Fourier space processing module receives the image preprocessed by the data interface and the preprocessing module, and sequentially performs Fast Fourier Transform (FFT), serialization, word segmentation, and embedding to convert it to K-space, obtaining Fourier space data. Then, it serializes the K-space data according to a preset strategy (such as frequency sorting or patch partitioning); performs tokenization to convert complex numerical data and positional information in the Fourier space data into tokens; implements data embedding, position embedding, and modal embedding layers, and outputs a token sequence acceptable to the Transformer model.

[0118] The Fourier Space Model Module - K-Space Transformer Model Module: This module contains a Transformer model based on an Encoder-Decoder structure, i.e., a deep learning model including an encoder and a decoder. This module receives the reference contrast sequence and the target low-frequency sequence from the Fourier Space Processing Module, i.e., the high-resolution reference space data and low-resolution target space data in the aforementioned embodiments. It performs encoding and decoding operations, outputting the predicted target high-frequency K-space token sequence, i.e., the sequence corresponding to the high-frequency region data. The reference contrast sequence is input to the Encoder, and the target low-frequency sequence is input to the Decoder.

[0119] The reconstruction and output module receives the sequence corresponding to the predicted high-frequency region data, i.e., the high-frequency token sequence. After K-space integration and data consistency, it performs an inverse fast Fourier transform to convert it back to complex Fourier space values. Specifically, it obtains low-resolution target space data, i.e., the original low-frequency data k_tar_LF, from the K-space processing module; performs K-space data integration (including data consistency steps); performs an inverse fast Fourier transform (IFFT) to convert the complete K-space sequence data back to the image domain, obtaining a high-resolution reconstructed image; and outputs or displays the results in an appropriate format (such as DICOM).

[0120] Model Management and Training Module (Offline Available): Responsible for loading pre-trained deep learning model parameters for inference, or, when a training set including labeled data is available, performing model training and parameter optimization according to the aforementioned model training methods.

[0121] In the aforementioned computer system, the input image is fed into the data interface and preprocessing module, then sequentially passes through the K-space processing module, the K-space Transformer model module, and the reconstruction and output module, until a high-resolution image is output. The encoder's output is then passed to the decoder's cross-attention layer.

[0122] This computer system can run as standalone software or be integrated into a PACS (Picture Archiving and Communication System), a medical image post-processing workstation, or an MRI equipment control console to provide rapid, high-quality MRI super-resolution reconstruction services for clinical diagnosis and research.

[0123] Based on the aforementioned computer system, a super-resolution method for magnetic resonance imaging is provided. Figure 6 This is a flowchart of the multi-contrast magnetic resonance imaging super-resolution method based on K-space Transformer provided in this specific embodiment. Figure 6 As shown, this multi-contrast magnetic resonance imaging super-resolution method includes the following steps:

[0124] Step S610: Input a multi-contrast image.

[0125] Specifically, a low-resolution T2-weighted image (I_T2_LR), i.e., the low-resolution target contrast magnetic resonance image in the aforementioned embodiment, and a high-resolution T1-weighted image (I_T1_HR) and a high-resolution FLAIR image (I_FLAIR_HR), i.e., the high-resolution reference contrast magnetic resonance image in the aforementioned embodiment, are input to the aforementioned data interface and preprocessing module.

[0126] Step S620: Convert to Fourier space.

[0127] Specifically, 2D Fast Fourier Transform (FFT) is performed on the three images to obtain Fourier spatial data k_T1_HR, k_FLAIR_HR including high-resolution reference spatial data, and k_T2_HR including high-resolution target spatial data. Then, the known low-frequency region data k_T2_LF in the high-resolution target spatial data k_T2_HR is determined. This known low-frequency region data in the high-resolution target spatial data refers to the data corresponding to the real low-resolution image in the aforementioned embodiment in K-space, i.e., the low-resolution target spatial data. The high-resolution target spatial data serves as the true value for calculating the loss function during the training of the transformer model. What needs to be generated is the unknown high-frequency region data corresponding to the high-resolution target spatial data.

[0128] Step S630, preprocessing and serialization.

[0129] Specifically, all K-space points of the high-resolution reference spatial data k_T1_HR and k_FLAIR_HR, as well as the K-space points of the low-resolution target spatial data k_T2_LF, are sorted according to their distance from the center of the K-space and divided into small patches (e.g., 4x4).

[0130] The complex values ​​within each patch are flattened and input into the embedding layer along with the patch's center frequency coordinates. Different modal embedding vectors are used to distinguish the tokens of T1, FLAIR, and T2_LF. The center frequency coordinates of the patch are encoded using learnable 2D positional embeddings.

[0131] Step S640: Input a deep learning model based on the encoding / decoding structure.

[0132] Specifically, the encoder is input with the token sequences corresponding to the high-resolution reference space data k_T1_HR and k_FLAIR_HR.

[0133] Decoder: The input low-resolution target space data, i.e., the token sequence corresponding to the low-frequency region data k_T2_LF in the high-resolution target space data, serves as the initial context. The decoder generates the predicted token sequence for the predicted high-frequency region data k_T2_HF by using self-attention on k_T2_LF and the generated portion, as well as cross-attention on the encoder output, in ascending frequency order (corresponding to the k_T2_HF region).

[0134] Further, Figure 7 This is a flowchart of obtaining the predicted target high-frequency Fourier spatial sequence provided in this specific embodiment, see reference. Figure 7The reference contrast Fourier space sequence and the target contrast low-frequency Fourier space sequence are respectively fed into the encoder and decoder of the deep learning model for position encoding, followed by a multi-head self-attention mechanism and an additive and normalization mechanism. Residuals are calculated after position encoding. Subsequently, the reference contrast Fourier space sequence is processed through a feedforward mechanism, and the encoder outputs a context representation. The target contrast low-frequency Fourier space sequence, processed by position encoding and multi-head self-attention, along with the encoder's output context representation, is then processed through a linear output processing unit via a multi-head cross-attention mechanism and a feedforward mechanism to output the predicted target high-frequency Fourier space sequence.

[0135] Step S650: Integration and reconstruction of magnetic resonance imaging super-resolution.

[0136] Specifically, the predicted high-frequency region data k_T2_HF_pred is merged with the known low-resolution target spatial data k_T2_LF to obtain the predicted high-resolution target spatial data k_T2_HR_pred. Then, a 2D inverse fast Fourier transform is performed on the predicted high-resolution target spatial data to obtain the reconstructed high-resolution T2 image I_T2_HR_pred, which is the super-resolution reconstructed target contrast magnetic resonance image.

[0137] Alternatively, the K-space L1 loss can be calculated using real high-frequency region data k_T2_HF_gt, and the image domain L1 loss can be added for joint optimization. For example, Loss_total = 1.0 × Loss_k_L1 + 0.1 × Loss_img_L1 can be set.

[0138] Alternatively, different Transformer variants, such as the Vision Transformer (ViT model), can be used to treat the K-space as an "image" for patch partitioning and processing. Different K-space ordering and tokenization strategies can be explored. This method can be extended to 3D MRI data. It can also be combined with other techniques, such as Generative Adversarial Networks (GANs), to further improve image quality.

[0139] The above specific embodiments provide a multi-contrast magnetic resonance imaging super-resolution method based on K-space Transformer, which regards the MRI super-resolution problem as an "information translation" process from source mode to target mode, which is performed in K-space and constrained by low-frequency information of target mode.

[0140] It has the following beneficial effects: Fully utilizes K-space information: Direct operation in K-space is more in line with the physical process of MRI imaging and can better utilize the frequency structure characteristics of K-space to generate high-frequency details.

[0141] Efficient multi-contrast information fusion: The Transformer's attention mechanism can capture long-distance and complex dependencies between K-space data of different contrasts, achieving deep and efficient information fusion, and using the information of the reference contrast to accurately guide the generation of high-frequency information of the target contrast.

[0142] Context constraint generation: Low-frequency information of the target contrast is used as the context input of the decoder to ensure that the generated high-frequency information is consistent with the low-frequency information, thereby improving the accuracy and robustness of reconstruction.

[0143] Powerful representation capabilities: By drawing on the powerful capabilities of Transformer in sequence modeling, it is expected to learn better data representations in MRI super-resolution reconstruction tasks, thereby reconstructing higher quality, more detailed high-resolution images.

[0144] Flexibility and scalability: The framework can easily handle different numbers and types of reference contrast inputs, offering good flexibility and scalability, and is suitable for various clinical multi-contrast MRI scanning protocols.

[0145] This embodiment also provides a magnetic resonance imaging super-resolution reconstruction apparatus, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. The terms "module," "unit," "subunit," etc., used below refer to combinations of software and / or hardware that perform a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0146] The magnetic resonance imaging super-resolution reconstruction device includes: an image acquisition module, a spatial transformation module, a sorting module, and a reconstruction module.

[0147] The image acquisition module is used to acquire at least one low-resolution target contrast magnetic resonance image corresponding to the super-resolution magnetic resonance imaging to be reconstructed, and at least one high-resolution reference contrast magnetic resonance image.

[0148] The spatial transformation module is used to transform the low-resolution target contrast magnetic resonance image and the high-resolution reference contrast magnetic resonance image from the original image space to the preset Fourier space to obtain the corresponding Fourier space data; the Fourier space data includes high-resolution reference space data and low-resolution target space data.

[0149] The sorting module is used to sort Fourier space data according to a preset sorting method to obtain Fourier space sequence data.

[0150] The reconstruction module is used to input Fourier space sequence data into the target deep learning model to obtain predicted high-frequency region data; the target deep learning model is trained based on a preset deep learning model with an encoding and decoding structure, and is used to generate high-frequency region data of target contrast in the preset Fourier space; it is also used to generate super-resolution reconstructed target contrast magnetic resonance images based on the predicted high-frequency region data and low-resolution target spatial data.

[0151] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.

[0152] This embodiment also provides an electronic device including a memory and a processor, the memory storing a computer program and the processor being configured to run the computer program to perform the steps in any of the above method embodiments.

[0153] 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.

[0154] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program:

[0155] S1, acquire at least one low-resolution target contrast magnetic resonance image corresponding to the super-resolution magnetic resonance imaging to be reconstructed, and at least one high-resolution reference contrast magnetic resonance image.

[0156] S2, convert the low-resolution target contrast magnetic resonance image and the high-resolution reference contrast magnetic resonance image from the original image space to the preset Fourier space to obtain the corresponding Fourier space data; the Fourier space data includes high-resolution reference space data and low-resolution target space data.

[0157] S3. Sort the Fourier space data according to the preset sorting method to obtain Fourier space sequence data.

[0158] S4, Fourier space sequence data is input into the target deep learning model to obtain predicted high-frequency region data; the target deep learning model is trained based on a preset deep learning model with an encoding and decoding structure, and is used to generate high-frequency region data of target contrast in the preset Fourier space.

[0159] S5 generates super-resolution reconstructed target contrast magnetic resonance images based on predicted high-frequency regional data and low-resolution target spatial data.

[0160] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated in this embodiment.

[0161] Furthermore, in conjunction with the magnetic resonance imaging super-resolution reconstruction method provided in the above embodiments, this embodiment can also provide a storage medium for implementation. The storage medium stores a computer program; when executed by a processor, the computer program implements any of the magnetic resonance imaging super-resolution reconstruction methods described in the above embodiments.

[0162] It should be understood that the specific embodiments described herein are merely illustrative of the application and not intended to limit it. All other embodiments derived by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0163] Obviously, the accompanying drawings are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar situations based on these drawings without any creative effort. Furthermore, it is understood that although the work done in this development process may be complex and lengthy, for those skilled in the art, certain design, manufacturing, or production modifications made based on the technical content disclosed in this application are merely conventional technical means and should not be considered as insufficient disclosure of this application.

[0164] The term "embodiment" in this application refers to a specific feature, structure, or characteristic described in connection with an embodiment that may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily imply the same embodiment, nor does it imply that it is mutually exclusive with or independent of other embodiments. It will be clearly or implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0165] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of patent protection. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the appended claims.

Claims

1. A method for super-resolution reconstruction using magnetic resonance imaging, characterized in that, The method includes: Acquire at least one low-resolution target contrast magnetic resonance image corresponding to the super-resolution magnetic resonance image to be reconstructed, and at least one registered high-resolution reference contrast magnetic resonance image; wherein, the target contrast refers to the type of magnetic resonance image to be super-resolution reconstructed; the reference contrast refers to a spatially registered high-resolution magnetic resonance image with different imaging contrast, used as supplementary information to guide the reconstruction of the target image. The low-resolution target contrast magnetic resonance image and the high-resolution reference contrast magnetic resonance image are respectively converted from the original image space to a preset Fourier space to obtain the corresponding Fourier space data; the Fourier space data includes high-resolution reference space data and low-resolution target space data. According to a preset sorting method, the Fourier space data is sorted to obtain Fourier space sequence data; wherein, the preset sorting method includes: expanding and sorting the spatial points corresponding to the Fourier space data in one dimension based on frequency, and also includes performing two-dimensional / three-dimensional serialization processing based on the blocks corresponding to the Fourier space data. The Fourier space sequence data is input into a target deep learning model to obtain predicted high-frequency region data. The target deep learning model is trained based on a preset deep learning model with an encoder-decoder structure and is used to generate high-frequency region data of target contrast in a preset Fourier space. The preset deep learning model includes an encoder and a decoder. The process of inputting the Fourier space sequence data into the target deep learning model to obtain predicted high-frequency region data includes: inputting the high-resolution reference space data processed according to the preset sorting method into the encoder to generate a context representation; inputting the context representation and the low-resolution target space data processed according to the preset sorting method into the decoder to generate predicted high-frequency region data. The decoder is used to generate a prediction sequence of predicted high-frequency region data by using known low-frequency region data in the high-resolution target space data and the generated predicted high-frequency region data, as well as the reference contrast context representation output by the encoder, in order from low to high frequency. Based on the predicted high-frequency region data and the low-resolution target space data, a super-resolution reconstructed target contrast magnetic resonance image is generated.

2. The magnetic resonance imaging super-resolution reconstruction method according to claim 1, characterized in that, The step of sorting the Fourier space data according to a preset sorting method to obtain Fourier space sequence data includes: Acquire the high-resolution reference space data and the low-resolution target space data, and the corresponding multiple spatial points; calculate the multiple spatial distances of the multiple spatial points from the center point in the preset Fourier space; Based on the multiple spatial distances, the multiple spatial points are arranged in a vector sequence, and based on a preset embedding strategy, one-dimensional sequence data in Fourier space is determined.

3. The magnetic resonance imaging super-resolution reconstruction method according to claim 1, characterized in that, The step of sorting the Fourier space data according to a preset sorting method to obtain Fourier space sequence data includes: According to the preset blocks, the high-resolution reference spatial data and the low-resolution target spatial data are divided into multiple spatial data blocks; Calculate the distance between the center point of the plurality of spatial data blocks and the center point of the plurality of spatial blocks in the preset Fourier space; Based on the distances between the multiple spatial blocks, the multiple spatial data blocks are arranged in a vector sequence, and based on a preset embedding strategy, the Fourier space multidimensional sequence data is determined.

4. The magnetic resonance imaging super-resolution reconstruction method according to any one of claim 2 or claim 3, characterized in that, The method further includes: The low-resolution target spatial data is determined according to a preset sampling strategy.

5. The magnetic resonance imaging super-resolution reconstruction method according to claim 4, characterized in that, The process of generating a super-resolution reconstructed target contrast magnetic resonance image based on the predicted high-frequency region data and the low-resolution target spatial data includes: The high-frequency region data and the low-resolution target space data are merged to obtain the predicted high-resolution target space data; The predicted high-resolution target spatial data is transformed from a preset Fourier space to the original image space to generate a super-resolution reconstructed target contrast magnetic resonance image.

6. The magnetic resonance imaging super-resolution reconstruction method according to claim 4, characterized in that, The method further includes: Extract high-frequency region data of targets from real high-resolution target training images in a preset training set; the preset training set also includes low-resolution target training images and high-resolution reference training images. The low-resolution target training image and the high-resolution reference training image are input into a preset deep learning model to determine the high-frequency region data for predicting the target contrast in Fourier space. Based on a preset loss function, the loss values ​​of the target high-frequency region data and the predicted high-frequency region data are determined; Based on the preset gradient descent algorithm and the loss value, the preset deep learning model is trained to obtain the target deep learning model.

7. A magnetic resonance imaging super-resolution reconstruction device, characterized in that, The device includes: an image acquisition module, a spatial transformation module, a sorting module, and a reconstruction module; The image acquisition module is used to acquire at least one low-resolution target contrast magnetic resonance image corresponding to the super-resolution magnetic resonance imaging to be reconstructed, and at least one registered high-resolution reference contrast magnetic resonance image; wherein, the target contrast refers to the type of magnetic resonance imaging to be super-resolution reconstructed; the reference contrast refers to a spatially registered high-resolution magnetic resonance imaging with different imaging contrast, used as supplementary information to guide the reconstruction of the target image. The spatial transformation module is used to transform the low-resolution target contrast magnetic resonance image and the high-resolution reference contrast magnetic resonance image from the original image space to a preset Fourier space, respectively, to obtain corresponding Fourier space data; the Fourier space data includes high-resolution reference space data and low-resolution target space data; The sorting module is used to sort the Fourier space data according to a preset sorting method to obtain Fourier space sequence data; wherein, the preset sorting method includes: expanding and sorting the spatial points corresponding to the Fourier space data in one dimension based on frequency, and also includes performing two-dimensional / three-dimensional serialization processing based on the blocks corresponding to the Fourier space data. The reconstruction module is used to input the Fourier space sequence data into a target deep learning model to obtain predicted high-frequency region data. The target deep learning model is trained based on a preset deep learning model with an encoder-decoder structure and is used to generate high-frequency region data of target contrast in a preset Fourier space. It is also used to generate a super-resolution reconstructed target contrast magnetic resonance image based on the predicted high-frequency region data and the low-resolution target space data. The preset deep learning model includes an encoder and a decoder. The step of inputting the Fourier space sequence data into the target deep learning model to obtain the predicted high-frequency region data includes: inputting the high-resolution reference space data processed according to the preset sorting method into the encoder to generate a context representation; inputting the context representation and the low-resolution target space data processed according to the preset sorting method into the decoder to generate the predicted high-frequency region data. The decoder is used to generate a predicted sequence of the predicted high-frequency region data in ascending order of frequency, using known low-frequency region data from the high-resolution target space data, the generated predicted high-frequency region data, and the reference contrast context representation output by the encoder.

8. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the magnetic resonance imaging super-resolution reconstruction method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the magnetic resonance imaging super-resolution reconstruction method according to any one of claims 1 to 6.

Images