A method, device and processing equipment for super-resolution magnetic resonance sodium imaging

Abstract

The present application provides a magnetic resonance sodium imaging image super-resolution method, apparatus, and processing equipment for configuring a novel super-resolution processing scheme, and thereby configuring a corresponding magnetic resonance sodium imaging image super-resolution model. This not only effectively improves the spatial resolution of magnetic resonance sodium imaging images, but also effectively increases the number of image layers, achieving high-quality magnetic resonance sodium imaging super-resolution effects, and can provide more comprehensive and detailed case information data support for related clinical work, thereby ensuring that the corresponding medical service quality is further improved.

Description

Technical Field

[0001] The present application relates to the field of magnetic resonance imaging, and in particular to a method, apparatus and processing equipment for super-resolution magnetic resonance sodium imaging. Background Art

[0002] In medical imaging, sodium magnetic resonance imaging (also known as sodium magnetic resonance imaging, 23 Sodium magnetic resonance imaging (MRI), a technique used to measure the distribution of sodium ions in the body, has made significant progress in recent years. MRI, which detects the distribution of sodium ions in the body, has been applied to the diagnosis of diseases such as osteoarthritis, multiple sclerosis, and brain tumors. In these conditions, MRI can provide relevant pathological information, helping doctors better understand the progression and impact of the disease.

[0003] However, the application of magnetic resonance sodium imaging faces a series of technical challenges. Sodium nuclei will produce four energy levels in a static magnetic field (main magnetic field), and there are three possible transition modes between these energy levels. 23 The interaction between the electric quadrupole moment of the sodium nucleus and the surrounding electric field gradient and macromolecules causes these transitions to decay rapidly, leading to a biexponential transverse relaxation phenomenon. Specifically, the relaxation time is relatively short, which directly affects the signal-to-noise ratio, resolution, and measurement accuracy of the image. Therefore, in clinical applications, due to the limitations of imaging equipment, the image resolution of magnetic resonance sodium imaging is often difficult to achieve ideal results.

[0004] To address this issue, image super-resolution technology was introduced. Image super-resolution technology aims to generate high-resolution images by processing low-resolution images, thereby improving image clarity and detail. In medical imaging, it can significantly improve image resolution, thereby improving the accuracy of disease diagnosis and reflecting more pathophysiological information.

[0005] However, existing super-resolution methods mainly focus on improving the resolution of a single image, that is, generating a high-resolution image from a single low-resolution image. This method is effective when processing multi-layer scan images, especially low-resolution CT images, but faces challenges in super-resolution magnetic resonance sodium imaging.

[0006] In magnetic resonance sodium imaging, since the number of low-resolution image layers obtained during the imaging process is relatively small, it becomes very difficult to generate high-resolution images with more layers. This is mainly because traditional super-resolution techniques assume that the number of layers of the input and output images is the same, but in practical applications, this assumption often does not hold true.

[0007] In other words, the general image super-resolution method in the existing technology is not suitable for the application scenario of magnetic resonance sodium imaging, and it is difficult to meet the image super-resolution requirements of magnetic resonance sodium imaging that take into account both high resolution and number of image layers. Summary of the Invention

[0008] The present application provides a magnetic resonance sodium imaging image super-resolution method, apparatus, and processing equipment for configuring a novel super-resolution processing scheme, and thereby configuring a corresponding magnetic resonance sodium imaging image super-resolution model. This not only effectively improves the spatial resolution of magnetic resonance sodium imaging images, but also effectively increases the number of image layers, achieving high-quality magnetic resonance sodium imaging super-resolution effects, and can provide more comprehensive and detailed case information data support for related clinical work, thereby ensuring that the corresponding medical service quality is further improved.

[0009] In a first aspect, the present application provides a method for super-resolution magnetic resonance sodium imaging, the method comprising:

[0010] acquiring an initial sodium magnetic resonance imaging image whose resolution is to be improved;

[0011] The initial magnetic resonance sodium imaging image is input into the magnetic resonance sodium imaging image super-resolution model, wherein the magnetic resonance sodium imaging image super-resolution model is used to improve the number of layers and resolution of the magnetic resonance sodium imaging image input into the model. The magnetic resonance sodium imaging image super-resolution model is pre-trained by the sample magnetic resonance sodium imaging image. During the working process of the magnetic resonance sodium imaging image super-resolution model, the magnetic resonance sodium imaging image input into the model is converted into a feature matrix F1 with a specification of w1, h1, n×64, where w represents width, h represents height, and n represents the number of layers. The feature matrix F1 is sampled into a feature matrix F2 of the same specification. The feature matrix after weighted processing of the feature matrix F2 is added to the feature matrix F1 to obtain a feature matrix F3. The feature matrix size of the feature matrix F3 is expanded to obtain a feature matrix F4 with a specification of w1×8, h1×8, n×8. The feature matrix F4 is down-sampled to obtain a feature matrix F2 with a specification of w2, h2, n×8×r 2 The characteristic matrix F5, r represents the scaling factor, w1×8=w2 / r 2 , h1×8=h2 / r 2 , convolve the feature matrix F5 to obtain the feature matrix F6 with specifications of w2,h2,n, and convert the feature matrix F6 into the output image;

[0012] Extract the target MRI image output by the MRI image super-resolution model.

[0013] In a second aspect, the present application provides a magnetic resonance sodium imaging super-resolution device, comprising:

[0014] an acquisition unit, configured to acquire an initial sodium magnetic resonance imaging image whose resolution is to be improved;

[0015] The input unit is used to input the initial magnetic resonance sodium imaging image into the magnetic resonance sodium imaging image super-resolution model, wherein the magnetic resonance sodium imaging image super-resolution model is used to improve the number of layers and resolution of the magnetic resonance sodium imaging image input by the model. The magnetic resonance sodium imaging image super-resolution model is pre-trained by the sample magnetic resonance sodium imaging image. During the working process of the magnetic resonance sodium imaging image super-resolution model, the magnetic resonance sodium imaging image input by the model is converted into a feature matrix F1 with a specification of w1, h1, n×64, where w represents width, h represents height, and n represents the number of layers. The feature matrix F1 is sampled into a feature matrix F2 of the same specification. The feature matrix after weighted processing of the feature matrix F2 is added to the feature matrix F1 to obtain a feature matrix F3. The feature matrix size of the feature matrix F3 is expanded to obtain a feature matrix F4 with a specification of w1×8, h1×8, n×8. The feature matrix F4 is down-sampled to obtain a feature matrix F2 with a specification of w2, h2, n×8×r 2 The characteristic matrix F5, r represents the scaling factor, w1×8=w2 / r 2 , h1×8=h2 / r 2 , convolve the feature matrix F5 to obtain the feature matrix F6 with specifications of w2,h2,n, and convert the feature matrix F6 into the output image;

[0016] The extraction unit is used to extract the target magnetic resonance sodium imaging image output by the magnetic resonance sodium imaging image super-resolution model.

[0017] In a third aspect, the present application provides a processing device comprising a processor and a memory, wherein a computer program is stored in the memory, and when the processor calls the computer program in the memory, the method provided in the first aspect of the present application or any possible implementation of the first aspect of the present application is executed.

[0018] In a fourth aspect, the present application provides a computer-readable storage medium, which stores multiple instructions, and the instructions are suitable for a processor to load to execute the method provided in the first aspect of the present application or any possible implementation of the first aspect of the present application.

[0019] From the above content, it can be concluded that this application has the following beneficial effects:

[0020] Aiming at the super-resolution goal of magnetic resonance sodium imaging, this application configures a novel super-resolution processing scheme, and configures a corresponding magnetic resonance sodium imaging image super-resolution model based on this. This can not only effectively improve the spatial resolution of magnetic resonance sodium imaging images, but also effectively increase the number of image layers, achieve high-quality magnetic resonance sodium imaging super-resolution effects, and provide more comprehensive and detailed case information data support for related clinical work, thereby ensuring that the corresponding medical service quality is further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 A schematic diagram of a flow chart of the magnetic resonance sodium imaging super-resolution method of the present application;

[0023] Figure 2 A schematic diagram of a scenario for the overall model processing architecture of this application;

[0024] Figure 3 A schematic diagram of a scenario for the depth upsampling block processing logic of this application;

[0025] Figure 4 A schematic diagram of a scenario for the dense feature extraction block processing logic of this application;

[0026] Figure 5 This is a schematic structural diagram of a magnetic resonance sodium imaging super-resolution device of the present application;

[0027] Figure 6 This is a structural diagram of the processing equipment for this application. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0029] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or modules is not necessarily limited to those steps or modules clearly listed, but may include other steps or modules that are not clearly listed or that are inherent to these processes, methods, products or devices. The naming or numbering of steps in this application does not mean that the steps in the method flow must be executed in the time / logical sequence indicated by the naming or numbering. The process steps that have been named or numbered can be changed in the execution order according to the technical purpose to be achieved, as long as the same or similar technical effects can be achieved.

[0030] The division of modules in this application is a logical division. In actual application, there may be other division methods. For example, multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection between modules can be electrical or other similar forms, which are not limited in this application. Moreover, the modules or submodules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed into multiple circuit modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this application.

[0031] Before introducing the magnetic resonance sodium imaging image super-resolution method provided by this application, the background content involved in this application is first introduced.

[0032] The magnetic resonance sodium imaging image super-resolution method, device and computer-readable storage medium provided in this application can be applied to processing equipment to configure a novel super-resolution processing scheme, and thereby configure a corresponding magnetic resonance sodium imaging image super-resolution model. This can not only effectively improve the spatial resolution of magnetic resonance sodium imaging images, but also effectively increase the number of image layers to achieve high-quality magnetic resonance sodium imaging super-resolution effects, and can provide more comprehensive and detailed case information data support for related clinical work, thereby ensuring that the corresponding medical service quality is further improved.

[0033] The magnetic resonance sodium imaging image super-resolution method mentioned in this application can be implemented by a magnetic resonance sodium imaging image super-resolution device, or by various types of processing devices such as a server, physical host, or user equipment (UE) that integrates the magnetic resonance sodium imaging image super-resolution device. The magnetic resonance sodium imaging image super-resolution device can be implemented using hardware or software, and the UE can specifically be a terminal device such as a smartphone, tablet computer, laptop computer, desktop computer, or personal digital assistant (PDA). The processing device can be set up in a device cluster.

[0034] It can be understood that in specific applications, the focus of the present application is to carry out corresponding magnetic resonance sodium imaging image super-resolution processing based on a pre-configured magnetic resonance sodium imaging image super-resolution model. In this case, the processing equipment that executes the magnetic resonance sodium imaging image super-resolution method of the present application or is equipped with the corresponding application service of the magnetic resonance sodium imaging image super-resolution method of the present application usually only needs to meet the required data processing capabilities. The specific equipment types and specific equipment deployment forms involved are relatively flexible and can be flexibly configured according to actual needs. This application does not make specific limitations.

[0035] If the training of a super-resolution model for magnetic resonance sodium imaging is further involved, further adaptive adjustments can be made based on actual conditions.

[0036] As an example, the processing device can be divided into two parts, that is, the processing device can include a first processing device that performs the training task of the magnetic resonance sodium imaging image super-resolution model and a second processing device that performs the application task of the magnetic resonance sodium imaging image super-resolution model.

[0037] Furthermore, in some cases, the present application solution may also involve further data support work such as lesion identification / localization to assist in disease diagnosis, and may further adaptively adjust the processing equipment corresponding to further data analysis and processing. Corresponding to the above example, in some cases, the processing equipment may also include a third processing device that performs data analysis and processing based on the super-resolution results of magnetic resonance sodium imaging images.

[0038] Furthermore, in some cases, the present application solution may also involve further image display, and the processing device itself may be configured with a display screen (including a touch screen), or an external display device, or other external device with a display screen to meet the requirements of the image display function.

[0039] Next, the magnetic resonance sodium imaging image super-resolution method provided by the present application is introduced.

[0040] First, see Figure 1 , Figure 1 A schematic flow chart of the magnetic resonance sodium imaging image super-resolution method of the present application is shown. The magnetic resonance sodium imaging image super-resolution method provided by the present application may specifically include the following steps S101 to S103:

[0041] Step S101, obtaining an initial magnetic resonance sodium imaging image whose resolution is to be improved;

[0042] It can be understood that, corresponding to the super-resolution requirement for the magnetic resonance sodium imaging image in actual situations, the magnetic resonance sodium imaging image with the currently required improved resolution or the resolution to be improved can be obtained.

[0043] For the convenience of explanation, the magnetic resonance sodium imaging images in different situations are respectively referred to as initial magnetic resonance sodium imaging images, sample magnetic resonance sodium imaging images, and target magnetic resonance sodium imaging images in this application.

[0044] Among them, magnetic resonance imaging technology itself is an existing concept. In actual situations, the magnetic resonance sodium imaging images obtained by scanning are usually in the form of image sequences. Considering that it is not the focus of this application, this application will not elaborate on it in detail.

[0045] As an example, the initial magnetic resonance sodium imaging image obtained here can be a low-resolution magnetic resonance sodium imaging image acquired in the sagittal position of the bFFE-UTE sequence, and the corresponding parameters are set as TR / TE=8.0 / 0.28ms, FOV=260×260×90mm, layer spacing of -3mm, and image sequence dimensions of (w,h,n)=(60,60,30), where w represents width, h represents height, and n represents the number of layers.

[0046] Among them, for the acquisition and processing of the initial magnetic resonance sodium imaging image, both manual entry and automatic acquisition can be used. The device can extract the image from the local storage space or the storage space of other devices, or the device can receive the image sent by other devices. This is all possible and corresponds to the flexible and changeable application needs in actual situations.

[0047] In addition, in actual applications, the solution of this application is usually processed in the form of a work task, that is, a super-resolution task of magnetic resonance sodium imaging. The initiation of the task can be done by manual entry or automatic initiation. The device can initiate it autonomously according to the preset autonomous initiation strategy / rules, or the device can receive tasks sent by other devices. This is similar to the above and meets the diverse application needs of the solution.

[0048] Step S102: Input the initial magnetic resonance sodium imaging image into a magnetic resonance sodium imaging image super-resolution model, wherein the magnetic resonance sodium imaging image super-resolution model is used to improve the number of layers and resolution of the magnetic resonance sodium imaging image input to the model. The magnetic resonance sodium imaging image super-resolution model is pre-trained by a sample magnetic resonance sodium imaging image. During the operation of the magnetic resonance sodium imaging image super-resolution model, the magnetic resonance sodium imaging image input to the model is converted into a feature matrix F1 with a specification of w1, h1, n×64, where w represents width, h represents height, and n represents the number of layers. The feature matrix F1 is sampled into a feature matrix F2 of the same specification. The feature matrix F2 is weighted and added to the feature matrix F1 to obtain a feature matrix F3. The feature matrix size of the feature matrix F3 is expanded to obtain a feature matrix F4 with a specification of w1×8, h1×8, and n×8. The feature matrix F4 is down-sampled to obtain a feature matrix F2 with a specification of w2, h2, n×8×r 2 The characteristic matrix F5, r represents the scaling factor, w1×8=w2 / r 2 , h1×8=h2 / r 2 , convolve the feature matrix F5 to obtain the feature matrix F6 with specifications of w2,h2,n, and convert the feature matrix F6 into the output image;

[0049] It can be seen that the present application designs a novel magnetic resonance sodium imaging image super-resolution logic / processing solution, which can be specifically implemented with the corresponding processing model as the carrier. In this regard, the present application can pre-configure the corresponding magnetic resonance sodium imaging image super-resolution model, which is trained by sample magnetic resonance sodium imaging images in the prior model training work / link.

[0050] Specifically, refer to Figure 2 A schematic diagram of a scenario of the overall model processing architecture of the present application is shown. The processing architecture of the magnetic resonance sodium imaging image super-resolution model (or asymmetric medical image super-resolution model) of the present application, or the asymmetric medical image super-resolution method designed by the present application, can have the following main contents:

[0051] 1) The convolution block performs preliminary image feature extraction and converts the image sequence (low-resolution image) input to the model into a matrix of size (w, h, n). In this way, the MRI image input to the model is converted into a feature matrix F1 (w1, h1, n×64), which corresponds to the input layer of the model.

[0052] 2) Feature sampling block performs feature enhancement and samples the feature matrix F1 into a feature matrix F2 of the same size (w1, h1, n×64);

[0053] 3) The self-attention mechanism introduced by self-attention gating is used to weight the feature matrix F2;

[0054] 4) In the matrix addition step corresponding to the self-attention mechanism, the feature matrix (w1, h1, n×64) after weighted processing of the feature matrix F2 is added to the feature matrix F1 to obtain the feature matrix F3 (w1, h1, n×64);

[0055] 5) The feature matrix size is expanded by the depth upsampling block, and the feature matrix size of the feature matrix F3 is expanded to obtain the feature matrix F4 (w1×8,h1×8,n×8);

[0056] 6) Downsample the feature matrix F4 to obtain the feature matrix F5 (w2,h2,n×8×r 2 ), w1×8=w2 / r 2 , h1×8=h2 / r 2 , r can be understood as an intermediate coefficient introduced in the parameter calculation process of this application;

[0057] 7) The convolution block convolves the feature matrix F5 to obtain the feature matrix F6 (w2, h2, n);

[0058] 8) Convert the feature matrix F6 into an output image (high-resolution image) as the model output. Specifically, this may involve the model structure of the corresponding model output link, such as the fully connected layer. Considering that it is not the focus of this application, and considering that the output layer itself is also a mature concept of neural networks in machine models, a specific explanation is not given.

[0059] Under the above-mentioned processing mechanism, it can be seen that this application combines the specific image specification configuration, and through up- and down-sampling, enlarges and restores the image to extract features, strengthen the original features, reduce the error of the generated layer image, and effectively achieve an increase in the number of image layers while improving the resolution.

[0060] In terms of details, the feature sampling block involved in this application, its processing logic or working process may include upsampling (on the input side), activation function and downsampling (on the output side) in sequence.

[0061] Correspondingly, as an exemplary embodiment, the feature sampling block samples the feature matrix F1 into a feature matrix F2 of the same specification, which may specifically include:

[0062] 2.1) The feature matrix F1 with the size of w1,h1,n×64 is sampled into the size of w×r,h×r,n×64 / r by the following formula 2 The characteristic matrix F11:

[0063] ,

[0064] As can be seen, this is the upsampling link involved in the feature sampling block. Starting from the quantization formula, a more specific and practical implementation solution is given.

[0065] Among them, mod is the modulus operator.

[0066] As an example, r, which acts as a scaling factor, can be 2, that is, r=2. In this case, the element value of each position in the magnified matrix X′ is calculated by the original matrix X based on the scaling factor r=2 using the above formula.

[0067] 2.2) Process the feature matrix F11 through the activation function ReLU to obtain the feature matrix F12;

[0068] Among them, ReLU, namely Linear rectification function, is itself a mature concept or common network structure in the neural network architecture. This application introduces it into the specific working logic / structure of the feature sampling block responsible for processing the feature matrix F1 into the feature matrix F2.

[0069] By inserting ReLU between the upsampling link above and the downsampling link below, in terms of specific implementation, it helps to improve the generalization and representation capabilities of the model.

[0070] 2.3) Downsample the feature matrix F12 into a feature matrix F2 of size w,h,n×64.

[0071] After downsampling, the feature matrix F1 is enhanced to obtain the feature matrix F2 that can be used for subsequent processing.

[0072] Furthermore, the processing logic or working process of the depth upsampling block involved in this application is as follows: Figure 3 A schematic diagram of a scenario of the depth upsampling block processing logic of the present application is shown, which may include, in sequence, a first dense feature extraction block (corresponding to the input side), a first convolution block, a second dense feature extraction block, a second convolution block, a third dense feature extraction block, a third convolution block, a first upsampling, a second upsampling, and a third upsampling (corresponding to the output side).

[0073] Correspondingly, as an exemplary embodiment, the feature matrix F3 is expanded by the depth upsampling block to obtain a feature matrix F4 with specifications of w1×8, h1×8, and n×8, which may specifically include:

[0074] The feature matrix F3 with the size of w1, h1, n×64 is processed by three deep feature extraction units in sequence, and then upsampled three times to obtain the feature matrix F4 with the size of w1×8, h1×8, n×8. The deep feature extraction unit includes one dense feature extraction block and one convolution block in sequence.

[0075] The feature matrix specifications before and after the dense feature extraction block remain unchanged (the changes in the feature matrix specifications before and after the deep feature extraction unit are mainly realized by the convolution block). The feature matrix specifications after the first deep feature extraction process are w1,h1,n×128, the feature matrix specifications after the second deep feature extraction process are w1,h1,n×256, and the feature matrix specifications after the third deep feature extraction process are w1,h1,n×512.

[0076] The feature matrix specifications after the first upsampling are w1×2, h1×2, n×128, the feature matrix specifications after the second upsampling are w1×4, h1×4, n×32, and the feature matrix specifications after the third upsampling are w1×8, h1×8, n×8.

[0077] It is easy to see that in the embodiment here, the present application designs a more complex set of processing results / logic for the depth upsampling processing involved in the depth upsampling block, combined with the specific image specification configuration, namely 3 groups of "dense feature extraction blocks + convolution blocks" + 3 groups of upsampling, so as to better achieve the processing effect of expanding the feature matrix F3 to obtain the feature matrix F4.

[0078] For the dense feature extraction block, the corresponding Figure 4 A schematic diagram of a scenario of the dense feature extraction block processing logic of the present application is shown as an exemplary embodiment, including:

[0079] In the dense feature extraction block, the input of the dense feature extraction block is connected to the input of the first convolution block, the second convolution block and the third convolution block respectively, the output of the first convolution block is connected to the input of the second convolution block and the input of the third convolution block respectively, the output of the second convolution block is connected to the input of the third convolution block, the input of the feature sampling block is connected to the input of the dense feature extraction block, the output of the first convolution block, the output of the second convolution block and the output of the third convolution block respectively, the input of the self-attention gate is connected to the feature sampling block, the input of the dense feature extraction block, the output of the first convolution block, the output of the second convolution block, the output of the third convolution block and the output of the self-attention gate are added as the output.

[0080] comparison Figure 4 , from the perspective of specific feature matrix processing, we can have:

[0081] 5.1) The input feature matrix F3 is passed through the first convolution block to extract features and obtain the feature matrix F31;

[0082] 5.2) After matrix addition of the feature matrix F3 and the feature matrix F31, the matrix is ​​input into the second convolution block for feature extraction to obtain the feature matrix F32;

[0083] 5.3) After matrix addition of the feature matrix F3, the feature matrix F31, and the feature matrix F32, the matrix is ​​input into the third convolution block for feature extraction to obtain the feature matrix F33;

[0084] 5.4) After matrix addition of feature matrix F3, feature matrix F31, feature matrix F32 and feature matrix F33, feature matrix F34 is obtained. Feature matrix F34 is further enhanced through feature sampling block and self-attention gating to obtain feature matrix F35;

[0085] Among them, the self-attention mechanism involved in the process of processing the feature matrix F34 into the feature matrix F35 can refer to the processing involved in the previous process of processing the feature matrix F2 into the feature matrix F3.

[0086] 5.5) Perform matrix addition on the feature matrix F34 and the feature matrix F35 to obtain the feature matrix F36 representing the deep features.

[0087] At this point, the feature matrix F36 can continue to improve the image resolution by upsampling three times.

[0088] The above content is an introduction to the specific model structure / working logic involved in the magnetic resonance sodium imaging image super-resolution model of this application. It is easy to understand that in specific applications, as mentioned above, it may involve prior model training processing. In this regard, the method of this application may also involve corresponding model training links.

[0089] Correspondingly, as an exemplary embodiment, before step S102 of inputting the initial magnetic resonance sodium imaging image into the magnetic resonance sodium imaging image super-resolution model, the method of the present application may further include:

[0090] Obtain sample sodium magnetic resonance imaging images and make annotations;

[0091] Based on sample magnetic resonance sodium imaging images, an initial model is trained to obtain a magnetic resonance sodium imaging image super-resolution model.

[0092] Among them, the labeling processing can be understood as configuring theoretical or standard model processing results for sample magnetic resonance sodium imaging images, that is, high-resolution magnetic resonance sodium imaging images, to assist the training model in the targeted super-resolution processing of magnetic resonance sodium imaging images during the model training process.

[0093] In specific operations, annotation processing can be done manually or by corresponding automatic annotation tools. When the automatic annotation tools are used in specific applications, the corresponding automatic annotation logic needs to be pre-configured.

[0094] As an example, the sample MRI image annotation may also specifically use an existing high-resolution image, such as a high-resolution MRI image acquired in the sagittal position using a bFFE-UTE sequence.

[0095] In this way, the obtained labeled sample magnetic resonance sodium imaging images can be used for specific model training.

[0096] The model training process usually includes the following:

[0097] In each model training stage, a training sample is input into the model, so that the model can carry out the corresponding magnetic resonance sodium imaging image super-resolution processing to realize forward propagation. Then, based on the magnetic resonance sodium imaging image super-resolution results output by the model, the loss function is calculated in combination with the annotation, and the model parameters are optimized according to the loss function calculation results to realize reverse propagation. After a large amount of training, if the preset model training requirements such as the number of training times, training time or prediction accuracy are met, the model training can be completed, and a magnetic resonance sodium imaging image super-resolution model that can be put into practical application can be obtained.

[0098] Among them, it can be understood that the specific model training architecture used in the training process and the specific loss function used in the training process can adopt the existing solution, or further optimize and improve the existing solution, or adopt a novel self-developed solution. These are all possible and can be configured according to actual conditions.

[0099] As an example, as a model training architecture, sample magnetic resonance sodium imaging images can be divided into training set, validation set and test set according to the ratio of 7:2:1 to be used for model training.

[0100] In addition, during the model training phase, other configuration tasks may also be involved, such as hyperparameters, model performance evaluation, etc. These can also be flexibly configured according to actual needs.

[0101] For model performance evaluation, this application can specifically use the existing metric Structural Similarity (SSIM) to evaluate the structural similarity of two images. The similarity between the model output image and the provided high-resolution image is calculated using three aspects: brightness, contrast, and structure. The corresponding specific formula is as follows:

[0102] ,

[0103] ,

[0104] in, and They are images The mean of all pixel values ​​in the matrix, is an image The covariance of and They are images The variance of all pixel values ​​in the matrix, b is the image depth, and the specific value can be 8.

[0105] The SSIM value range is [-1, 1]. The larger the value, the more similar it is. The performance of the model is evaluated based on this.

[0106] In addition, in order to further improve the model training effect, the present application can also continue to perform secondary processing on the sample magnetic resonance sodium imaging images to obtain a richer sample volume.

[0107] Correspondingly, as an exemplary embodiment, the aforementioned acquisition of the sample magnetic resonance sodium imaging image may specifically include:

[0108] Acquiring an initial sample sodium magnetic resonance imaging image, wherein the initial sample sodium magnetic resonance imaging image is an image sequence;

[0109] According to the preset number of high-resolution image layers corresponding to the model output, adjacent layers of each low-resolution image in the initial sample magnetic resonance sodium imaging image are interpolated. If k-1 layers of images are inserted between adjacent layers, the original adjacent layer image matrices are added and multiplied by 1 / k, 2 / k, ..., (k-1) / k respectively to obtain k-1 matrices. The k-1 matrices are converted into images to obtain a target sample magnetic resonance sodium imaging image with the preset number of high-resolution image layers as the sample magnetic resonance sodium imaging image, where k-1 is the number of interpolation layers.

[0110] It can be understood that in the embodiment herein, the sample secondary processing designed in the present application is based on interpolation to effectively further improve and expand the number of image layers.

[0111] The above content can be better understood by combining it with the following set of examples involved in practical applications.

[0112] S1: bFFE-UTE sequence was used to acquire low-resolution and corresponding high-resolution sagittal images. 23Na MRI sequence, acquisition parameters set to TR / TE = 8.0 / 0.28ms, FOV = 260×260×90mm, low-resolution and high-resolution image interslice spacing of -3mm and -1.5mm, respectively. The acquired data was divided into training, validation, and test sets according to a 7:2:1 ratio. The low-resolution image sequence dimension was (w,h,n)=(60,60,30), indicating an image width × height of 60×60, 30 slices, and each slice was an 8-bit single-channel image. The high-resolution image sequence dimension was (80,80,59);

[0113] S2: Insert a layer between adjacent layers of images in each low-resolution image sequence. The matrix of the newly inserted image is the matrix addition of the adjacent layer image matrices, multiplied by 1 / 2, and finally the low-resolution image matrix is ​​expanded from 30 layers to 59 layers.

[0114] S3: Construct a super-resolution model for sodium magnetic resonance imaging. Use the training set for model training. During the model training process, use the validation set to tune the model parameters and obtain the model with the best AUC on the validation set. The image processing flow of the model is as follows:

[0115] S31: The expanded image sequence matrix is ​​transformed into a feature matrix through a convolution operation to obtain F1. The size of F1 is (60, 60, 59 × 64). F1 is enhanced through a feature sampling block to obtain F2. The size of F2 is still (60, 60, 59 × 64). The specific processing of the feature sampling block is as follows:

[0116] S311: Amplify the information through the upsampling algorithm. The matrix size after F1 amplification is (60×2, 60×2, 59×16);

[0117] S312: Use the activation function ReLU to process the transformed matrix to improve the generalization and representation capabilities of the model;

[0118] S313: Perform the inverse upsampling operation on the amplified information through downsampling to obtain F2, where the size of F2 is (60, 60, 59×64);

[0119] S32: After weighted processing of F2 through the self-attention module, matrix addition is performed with F1 to obtain F3;

[0120] S33: The feature matrix size of F3 is expanded through the depth upsampling block to obtain F4 of size (60×8, 60×8, 59×8). F4 is downsampled to obtain F5 of size (80, 80, 59×8×36). F5 is further convolved to obtain F6 of size (80, 80, 59). It is converted into an image to obtain the final high-resolution image sequence. The processing of the depth upsampling block is as follows:

[0121] S331: Deep feature extraction is performed using three groups of "dense feature extraction blocks + convolution blocks." The feature matrix size after the first group is (60, 60, 59 × 128), the feature matrix size after the second group is (60, 60, 59 × 256), and the feature matrix size after the third group is (60, 60, 59 × 512). The matrix size remains unchanged before and after the dense feature extraction block processing. Taking the dense feature extraction block in the first group as an example, the specific processing flow is as follows:

[0122] S3311: extract the features of input F3 through the first convolution block to obtain F31;

[0123] S3312: After matrix addition of F3 and F31, the matrix is ​​input into the second convolution block for feature extraction to obtain F32;

[0124] S3313: After matrix addition of F3, F31, and F32, the matrix is ​​input into the third convolution block for feature extraction to obtain F33;

[0125] S3314: Matrix addition of F3, F31, F32, and F33 is performed to obtain F34, which is then enhanced using feature sampling blocks and self-attention gating to obtain F35.

[0126] S3315: Perform matrix addition on F34 and F35 to obtain F36.

[0127] S332: Upsample F36 three times to improve the image resolution and obtain matrix F4. After the first upsampling, the feature matrix size is (60×2, 60×2, 59×128), after the second upsampling, the feature matrix size is (60×4, 60×4, 59×32), and after the third upsampling, the feature matrix size is (60×8, 60×8, 59×8).

[0128] Step S103 : extracting the target magnetic resonance sodium imaging image output by the magnetic resonance sodium imaging image super-resolution model.

[0129] After the current initial magnetic resonance sodium imaging image is processed by the magnetic resonance sodium imaging image super-resolution model to obtain the corresponding target magnetic resonance sodium imaging image with effectively improved resolution and number of layers, the model will output it.

[0130] Correspondingly, the target magnetic resonance sodium imaging image output by the magnetic resonance sodium imaging image super-resolution model can be extracted.

[0131] At this point, it can be understood that for the target magnetic resonance sodium imaging image, which is the magnetic resonance sodium imaging image super-resolution result, local storage, remote storage, result forwarding, output of completion processing prompts, result display, or further data processing and analysis (such as the lesion identification / localization and other data support work for auxiliary disease diagnosis mentioned above), etc. Obviously, the specific data application content involved is relatively flexible and can be adjusted according to the pre- and real-time configured data application strategies / rules, and this application does not make specific limitations.

[0132] Finally, regarding the above solution content, in general, for the super-resolution goal of magnetic resonance sodium imaging, this application configures a novel super-resolution processing solution, and configures the corresponding magnetic resonance sodium imaging image super-resolution model based on this. This can not only effectively improve the spatial resolution of magnetic resonance sodium imaging images, but also effectively increase the number of image layers, achieve high-quality magnetic resonance sodium imaging super-resolution effects, and provide more comprehensive and detailed case information data support for related clinical work, thereby ensuring that the corresponding medical service quality is further improved.

[0133] The above is an introduction to the magnetic resonance sodium imaging image super-resolution method provided in this application. In order to facilitate better implementation of the magnetic resonance sodium imaging image super-resolution method provided in this application, this application also provides a magnetic resonance sodium imaging image super-resolution device from the perspective of functional modules.

[0134] See Figure 5 , Figure 5 This is a schematic diagram of the structure of the magnetic resonance sodium imaging image super-resolution device of the present application. In the present application, the magnetic resonance sodium imaging image super-resolution device 500 may specifically include the following structure:

[0135] An acquisition unit 501 is configured to acquire an initial sodium magnetic resonance imaging image whose resolution is to be improved;

[0136] An input unit 502 is configured to input an initial MRI image into a MRI super-resolution model, wherein the MRI super-resolution model is configured to improve the number of layers and resolution of the MRI image input to the model. The MRI super-resolution model is pre-trained using sample MRI images. During operation of the MRI super-resolution model, the MRI image input to the model is converted into a feature matrix F1 having a size of w1, h1, n×64, where w represents width, h represents height, and n represents the number of layers. The feature matrix F1 is sampled into a feature matrix F2 having the same size. The feature matrix F2 is weighted and added to the feature matrix F1 to obtain a feature matrix F3. The feature matrix F3 is expanded to obtain a feature matrix F4 having a size of w1×8, h1×8, and n×8. The feature matrix F4 is down-sampled to obtain a feature matrix F2 having a size of w2, h2, n×8×r 2 The characteristic matrix F5, r represents the scaling factor, w1×8=w2 / r 2 , h1×8=h2 / r 2 , convolve the feature matrix F5 to obtain the feature matrix F6 with specifications of w2,h2,n, and convert the feature matrix F6 into the output image;

[0137] The extraction unit 503 is configured to extract the target magnetic resonance sodium imaging image output by the magnetic resonance sodium imaging image super-resolution model.

[0138] In an exemplary embodiment, sampling the feature matrix F1 into a feature matrix F2 of the same specification specifically includes:

[0139] The feature matrix F1 with the size of w1,h1,n×64 is sampled into the size of w×r,h×r,n×64 / r by the following formula: 2 The characteristic matrix F11:

[0140] ,

[0141] in, Indicates output, Represents input, mod represents the modulo operator;

[0142] The feature matrix F11 is processed by the activation function ReLU to obtain the feature matrix F12;

[0143] The feature matrix F12 is downsampled into a feature matrix F2 with a size of w,h,n×64.

[0144] In another exemplary embodiment, the feature matrix F3 is enlarged to obtain a feature matrix F4 with specifications of w1×8, h1×8, and n×8, including:

[0145] The feature matrix F3 with the size of w1, h1, n×64 is processed by three deep feature extraction units in sequence, and then upsampled three times to obtain the feature matrix F4 with the size of w1×8, h1×8, n×8. The deep feature extraction unit includes one dense feature extraction block and one convolution block in sequence.

[0146] The feature matrix specifications before and after the dense feature extraction block processing remain unchanged. The feature matrix specifications after the first deep feature extraction processing are w1,h1,n×128, the feature matrix specifications after the second deep feature extraction processing are w1,h1,n×256, and the feature matrix specifications after the third deep feature extraction processing are w1,h1,n×512;

[0147] The feature matrix specifications after the first upsampling are w1×2, h1×2, n×128, the feature matrix specifications after the second upsampling are w1×4, h1×4, n×32, and the feature matrix specifications after the third upsampling are w1×8, h1×8, n×8.

[0148] In another exemplary embodiment, in the dense feature extraction block, the input of the dense feature extraction block is respectively connected to the inputs of the first convolution block, the second convolution block and the third convolution block, the output of the first convolution block is respectively connected to the input of the second convolution block and the input of the third convolution block, the output of the second convolution block is connected to the input of the third convolution block, the input of the feature sampling block is respectively connected to the input of the dense feature extraction block, the output of the first convolution block, the output of the second convolution block and the output of the third convolution block, the input of the self-attention gate is connected to the feature sampling block, and the input of the dense feature extraction block, the output of the first convolution block, the output of the second convolution block, the output of the third convolution block and the output of the self-attention gate are added as the output.

[0149] In another exemplary embodiment, the apparatus further includes a training unit 504 configured to:

[0150] Obtain sample sodium magnetic resonance imaging images and make annotations;

[0151] Based on sample magnetic resonance sodium imaging images, an initial model is trained to obtain a magnetic resonance sodium imaging image super-resolution model.

[0152] In another exemplary embodiment, the acquiring unit 501 is specifically configured to:

[0153] Acquiring an initial sample sodium magnetic resonance imaging image, wherein the initial sample sodium magnetic resonance imaging image is an image sequence;

[0154] According to the preset number of high-resolution image layers corresponding to the model output, adjacent layers of each low-resolution image in the initial sample magnetic resonance sodium imaging image are interpolated. If k-1 layers of images are inserted between adjacent layers, the original adjacent layer image matrices are added and multiplied by 1 / k, 2 / k, ..., (k-1) / k respectively to obtain k-1 matrices. The k-1 matrices are converted into images to obtain a target sample magnetic resonance sodium imaging image with the preset number of high-resolution image layers as the sample magnetic resonance sodium imaging image, where k-1 is the number of interpolation layers.

[0155] In another exemplary embodiment, the device further includes a display unit 505, configured to:

[0156] Display target magnetic resonance sodium imaging image.

[0157] This application also provides a processing device from the perspective of hardware structure, see Figure 6 , Figure 6 The schematic diagram of the structure of the processing device of the present application is shown. Specifically, the processing device of the present application may include a processor 601, a memory 602 and an input / output device 603. The processor 601 is used to execute the computer program stored in the memory 602 to implement the following Figure 1 The steps of the method for super-resolution magnetic resonance sodium imaging in the corresponding embodiment; or, the processor 601 is used to execute the computer program stored in the memory 602 to implement the following Figure 5 The memory 602 is used to store the functions of each unit in the embodiment corresponding to the processor 601. Figure 1 The computer program required for the magnetic resonance sodium imaging super-resolution method in the corresponding embodiment.

[0158] For example, the computer program may be divided into one or more modules / units, one or more of which are stored in the memory 602 and executed by the processor 601 to complete the present application. One or more modules / units may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in a computer device.

[0159] The processing device may include, but is not limited to, a processor 601, a memory 602, and an input / output device 603. Those skilled in the art will appreciate that the illustrations are merely examples of processing devices and do not limit the processing device. The processing device may include more or fewer components than shown, or a combination of certain components, or different components. For example, the processing device may also include a network access device, a bus, etc., and the processor 601, the memory 602, the input / output device 603, etc. are connected via a bus.

[0160] The processor 601 may be a central processing unit (CPU), or other general-purpose processors, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor. The processor is the control center of the processing device and connects various parts of the entire device using various interfaces and lines.

[0161] Memory 602 can be used to store computer programs and / or modules. Processor 601 implements various functions of the computer device by running or executing computer programs and / or modules stored in memory 602 and accessing data stored in memory 602. Memory 602 may primarily include a program storage area and a data storage area. The program storage area may store an operating system, at least one application required for a function, and the data storage area may store data generated based on the use of the processing device. Furthermore, memory may include high-speed random access memory (RAM) and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0162] When the processor 601 is used to execute the computer program stored in the memory 602, it can specifically implement the following functions:

[0163] acquiring an initial sodium magnetic resonance imaging image whose resolution is to be improved;

[0164] The initial magnetic resonance sodium imaging image is input into the magnetic resonance sodium imaging image super-resolution model, wherein the magnetic resonance sodium imaging image super-resolution model is used to improve the number of layers and resolution of the magnetic resonance sodium imaging image input into the model. The magnetic resonance sodium imaging image super-resolution model is pre-trained by the sample magnetic resonance sodium imaging image. During the working process of the magnetic resonance sodium imaging image super-resolution model, the magnetic resonance sodium imaging image input into the model is converted into a feature matrix F1 with a specification of w1, h1, n×64, where w represents width, h represents height, and n represents the number of layers. The feature matrix F1 is sampled into a feature matrix F2 of the same specification. The feature matrix after weighted processing of the feature matrix F2 is added to the feature matrix F1 to obtain a feature matrix F3. The feature matrix size of the feature matrix F3 is expanded to obtain a feature matrix F4 with a specification of w1×8, h1×8, n×8. The feature matrix F4 is down-sampled to obtain a feature matrix F2 with a specification of w2, h2, n×8×r 2 The characteristic matrix F5, r represents the scaling factor, w1×8=w2 / r 2 , h1×8=h2 / r 2 , convolve the feature matrix F5 to obtain the feature matrix F6 with specifications of w2,h2,n, and convert the feature matrix F6 into the output image;

[0165] Extract the target MRI image output by the MRI image super-resolution model.

[0166] Those skilled in the art will clearly understand that for the convenience and brevity of description, the specific working process of the magnetic resonance sodium imaging image super-resolution device, processing equipment and corresponding units described above can refer to the following. Figure 1 The description of the magnetic resonance sodium imaging image super-resolution method in the corresponding embodiment will not be repeated here.

[0167] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be accomplished by instructions, or by controlling related hardware through instructions. The instructions may be stored in a computer-readable storage medium and loaded and executed by a processor.

[0168] To this end, the present application provides a computer-readable storage medium, which stores a plurality of instructions, which can be loaded by a processor to execute the present application as follows: Figure 1 The steps of the magnetic resonance sodium imaging image super-resolution method in the corresponding embodiment, the specific operations can be referred to as follows Figure 1 The description of the magnetic resonance sodium imaging image super-resolution method in the corresponding embodiment will not be repeated here.

[0169] The computer-readable storage medium may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0170] Due to the instructions stored in the computer readable storage medium, the present application can be executed as follows: Figure 1 The steps of the magnetic resonance sodium imaging image super-resolution method in the corresponding embodiment can thus be implemented as follows: Figure 1 The beneficial effects that can be achieved by the magnetic resonance sodium imaging image super-resolution method in the corresponding embodiment are detailed in the previous description and will not be repeated here.

[0171] The above is a detailed introduction to the magnetic resonance sodium imaging image super-resolution method, apparatus, processing equipment, and computer-readable storage medium provided by the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the core idea of ​​the present application. At the same time, for those skilled in the art, based on the ideas of the present application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present application.

Claims

1. A method for super-resolution magnetic resonance sodium imaging, characterized in that: The method comprises: acquiring an initial sodium magnetic resonance imaging image whose resolution is to be improved; The initial magnetic resonance sodium imaging image is input into a magnetic resonance sodium imaging image super-resolution model, wherein the magnetic resonance sodium imaging image super-resolution model is used to improve the number of layers and resolution of the magnetic resonance sodium imaging image input into the model, and the magnetic resonance sodium imaging image super-resolution model is pre-trained by a sample magnetic resonance sodium imaging image. During the operation of the magnetic resonance sodium imaging image super-resolution model, the magnetic resonance sodium imaging image input into the model is converted into a feature matrix F1 with a specification of w1, h1, n×64, where w represents width, h represents height, and n represents the number of layers. The feature matrix F1 is sampled into a feature matrix F2 of the same specification, and the feature matrix after weighted processing of the feature matrix F2 is added to the feature matrix F1 to obtain a feature matrix F3. The feature matrix size of the feature matrix F3 is expanded to obtain a feature matrix F4 with a specification of w1×8, h1×8, n×8. The feature matrix F4 is down-sampled to obtain a feature matrix F2 with a specification of w2, h2, n×8×r 2 The characteristic matrix F5, r represents the scaling factor, w1×8=w2 / r 2 , h1×8=h2 / r 2 , convolve the feature matrix F5 to obtain a feature matrix F6 with a specification of w2,h2,n, and convert the feature matrix F6 into an output image; Extracting a target magnetic resonance sodium imaging image output by the magnetic resonance sodium imaging image super-resolution model; Sampling the feature matrix F1 into the feature matrix F2 of the same specification specifically includes: The feature matrix F1 with the size of w1,h1,n×64 is sampled into the size of w×r,h×r,n×64 / r by the following formula: 2 The characteristic matrix F11: ， in, Indicates output, Represents input, mod represents the modulo operator; The feature matrix F11 is processed by the activation function ReLU to obtain the feature matrix F12; Downsampling the feature matrix F12 into the feature matrix F2 with a size of w,h,n×64; The characteristic matrix F3 is enlarged to obtain the characteristic matrix F4 with specifications of w1×8, h1×8, and n×8, including: The feature matrix F3 with a size of w1, h1, n×64 is sequentially subjected to deep feature extraction processing by three deep feature extraction units, and then upsampled three times to obtain the feature matrix F4 with a size of w1×8, h1×8, n×8, wherein the deep feature extraction unit sequentially includes one dense feature extraction block and one convolution block; The feature matrix specifications before and after the dense feature extraction block processing remain unchanged. The feature matrix specifications after the first deep feature extraction processing are w1,h1,n×128, the feature matrix specifications after the second deep feature extraction processing are w1,h1,n×256, and the feature matrix specifications after the third deep feature extraction processing are w1,h1,n×512; The feature matrix specifications after the first upsampling are w1×2, h1×2, n×128, the feature matrix specifications after the second upsampling are w1×4, h1×4, n×32, and the feature matrix specifications after the third upsampling are w1×8, h1×8, n×8; In the dense feature extraction block, the input of the dense feature extraction block is respectively connected to the inputs of the first convolution block, the second convolution block and the third convolution block, the output of the first convolution block is respectively connected to the input of the second convolution block and the input of the third convolution block, the output of the second convolution block is connected to the input of the third convolution block, the input of the feature sampling block is respectively connected to the input of the dense feature extraction block, the output of the first convolution block, the output of the second convolution block and the output of the third convolution block, the input of the self-attention gate is connected to the feature sampling block, and the input of the dense feature extraction block, the output of the first convolution block, the output of the second convolution block, the output of the third convolution block and the output of the self-attention gate are added as the output; Before inputting the initial magnetic resonance sodium imaging image into the magnetic resonance sodium imaging image super-resolution model, the method further includes: Obtaining a sodium magnetic resonance imaging image of the sample and annotating it; Based on the sample magnetic resonance sodium imaging image, an initial model is trained to obtain the magnetic resonance sodium imaging image super-resolution model; Acquiring a magnetic resonance sodium imaging image of the sample, comprising: Acquiring an initial sample sodium magnetic resonance imaging image, wherein the initial sample sodium magnetic resonance imaging image is an image sequence; According to the preset number of high-resolution image layers corresponding to the model output, adjacent layers of each low-resolution image in the initial sample magnetic resonance sodium imaging image are interpolated. If k-1 layers of images are inserted between adjacent layers, the original adjacent layer image matrices are matrix added and then multiplied by 1 / k, 2 / k, ..., (k-1) / k respectively to obtain k-1 matrices, and the k-1 matrices are converted into images to obtain a target sample magnetic resonance sodium imaging image with the preset number of high-resolution image layers as the sample magnetic resonance sodium imaging image, where k-1 is the number of interpolation layers.

2. The method according to claim 1, characterized in that The method further comprises: The target magnetic resonance sodium imaging image is displayed.

3. A magnetic resonance sodium imaging image super-resolution device, characterized in that: The device comprises: an acquisition unit, configured to acquire an initial sodium magnetic resonance imaging image whose resolution is to be improved; An input unit is configured to input the initial magnetic resonance sodium imaging image into a magnetic resonance sodium imaging image super-resolution model, wherein the magnetic resonance sodium imaging image super-resolution model is configured to improve the number of layers and resolution of the magnetic resonance sodium imaging image input into the model, and the magnetic resonance sodium imaging image super-resolution model is pre-trained by a sample magnetic resonance sodium imaging image. During the operation of the magnetic resonance sodium imaging image super-resolution model, the magnetic resonance sodium imaging image input into the model is converted into a feature matrix F1 with a specification of w1, h1, n×64, where w represents width, h represents height, and n represents the number of layers. The feature matrix F1 is sampled into a feature matrix F2 with the same specification, a feature matrix after weighted processing of the feature matrix F2 is added to the feature matrix F1 to obtain a feature matrix F3, the feature matrix size of the feature matrix F3 is expanded to obtain a feature matrix F4 with a specification of w1×8, h1×8, n×8, and the feature matrix F4 is down-sampled to obtain a feature matrix F2 with a specification of w2, h2, n×8×r 2 The characteristic matrix F5, r represents the scaling factor, w1×8=w2 / r 2 , h1×8=h2 / r 2 , convolve the feature matrix F5 to obtain a feature matrix F6 with a specification of w2,h2,n, and convert the feature matrix F6 into an output image; an extraction unit, configured to extract a target magnetic resonance sodium imaging image output by the magnetic resonance sodium imaging image super-resolution model; Sampling the feature matrix F1 into the feature matrix F2 of the same specification specifically includes: The feature matrix F1 with the size of w1,h1,n×64 is sampled into the size of w×r,h×r,n×64 / r by the following formula: 2 The characteristic matrix F11: ， in, Indicates output, Represents input, mod represents the modulo operator; The feature matrix F11 is processed by the activation function ReLU to obtain the feature matrix F12; Downsampling the feature matrix F12 into the feature matrix F2 with a size of w,h,n×64; The characteristic matrix F3 is enlarged to obtain the characteristic matrix F4 with specifications of w1×8, h1×8, and n×8, including: The feature matrix F3 with a size of w1, h1, n×64 is sequentially subjected to deep feature extraction processing by three deep feature extraction units, and then upsampled three times to obtain the feature matrix F4 with a size of w1×8, h1×8, n×8, wherein the deep feature extraction unit sequentially includes one dense feature extraction block and one convolution block; The feature matrix specifications before and after the dense feature extraction block processing remain unchanged. The feature matrix specifications after the first deep feature extraction processing are w1,h1,n×128, the feature matrix specifications after the second deep feature extraction processing are w1,h1,n×256, and the feature matrix specifications after the third deep feature extraction processing are w1,h1,n×512; The feature matrix specifications after the first upsampling are w1×2, h1×2, n×128, the feature matrix specifications after the second upsampling are w1×4, h1×4, n×32, and the feature matrix specifications after the third upsampling are w1×8, h1×8, n×8; In the dense feature extraction block, the input of the dense feature extraction block is respectively connected to the inputs of the first convolution block, the second convolution block and the third convolution block, the output of the first convolution block is respectively connected to the input of the second convolution block and the input of the third convolution block, the output of the second convolution block is connected to the input of the third convolution block, the input of the feature sampling block is respectively connected to the input of the dense feature extraction block, the output of the first convolution block, the output of the second convolution block and the output of the third convolution block, the input of the self-attention gate is connected to the feature sampling block, and the input of the dense feature extraction block, the output of the first convolution block, the output of the second convolution block, the output of the third convolution block and the output of the self-attention gate are added as the output; The apparatus further comprises a training unit, configured to: Obtaining a sodium magnetic resonance imaging image of the sample and annotating it; Based on the sample magnetic resonance sodium imaging image, an initial model is trained to obtain the magnetic resonance sodium imaging image super-resolution model; The training unit is specifically used to: Acquiring an initial sample sodium magnetic resonance imaging image, wherein the initial sample sodium magnetic resonance imaging image is an image sequence; According to the preset number of high-resolution image layers corresponding to the model output, adjacent layers of each low-resolution image in the initial sample magnetic resonance sodium imaging image are interpolated. If k-1 layers of images are inserted between adjacent layers, the original adjacent layer image matrices are matrix added and then multiplied by 1 / k, 2 / k, ..., (k-1) / k respectively to obtain k-1 matrices, and the k-1 matrices are converted into images to obtain a target sample magnetic resonance sodium imaging image with the preset number of high-resolution image layers as the sample magnetic resonance sodium imaging image, where k-1 is the number of interpolation layers.

4. A processing device, characterized in that The method comprises a processor and a memory, wherein a computer program is stored in the memory, and when the processor calls the computer program in the memory, the method according to claim 1 or 2 is executed.

5. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute the method according to claim 1 or 2.

Images