Magnetic resonance image contrast conversion method and device, medium and product

Abstract

The invention discloses a magnetic resonance image contrast conversion method and device, a medium and a product, and relates to the technical field of magnetic resonance image processing algorithms, and the method comprises the steps: obtaining multi-contrast magnetic resonance image data comprising three adjacent slices, inputting the data into a magnetic resonance image contrast conversion model, and obtaining a target contrast magnetic resonance image; the determination process of the model comprises the steps that a magnetic resonance image sample data set is constructed, and each sample comprises a sample and magnetic resonance image data of true value contrast; constructing a magnetic resonance image contrast conversion network and a discriminator network; inputting each sample into a magnetic resonance image contrast conversion network to obtain a predicted contrast magnetic resonance image, and inputting the samples, the predicted contrast magnetic resonance image and the true contrast magnetic resonance image into a discriminator network to obtain a discriminator result; and a loss function is calculated to update a magnetic resonance image contrast conversion model, so that the quality of the magnetic resonance image after contrast conversion is effectively improved.

Description

Technical Field

[0001] The present application relates to the technical field of magnetic resonance image processing algorithms, and in particular to a magnetic resonance image contrast conversion method, device, medium and product. Background Art

[0002] Multi-contrast Magnetic Resonance Imaging (MRI) plays a key role in the comprehensive evaluation of human anatomy and function. As a complementary tool, it not only improves diagnostic accuracy but also enhances the effectiveness of subsequent imaging tasks by obtaining diverse tissue information from different contrasts. However, the widespread application of multi-contrast MRI remains elusive due to its high cost and time. In addition, for developing infants, there are significant challenges in acquiring high-quality multi-contrast MRI data due to their irregular sleep patterns and limited attention span, which makes it difficult for them to follow instructions and remain still. MRI image conversion technology provides an effective solution that overcomes this challenge by synthesizing the missing target contrast under the guidance of the existing source contrast.

[0003] However, existing MRI image conversion technology mainly focuses on single contrast and single slice data, that is, it can only realize the conversion of single source contrast MRI images to single target contrast MRI images, and only considers the data of single slice MRI images, which makes the converted MRI image results have the problem of poor imaging quality. Summary of the invention

[0004] The purpose of the present application is to provide a magnetic resonance image contrast conversion method, device, medium and product, which can effectively improve the quality of magnetic resonance images after contrast conversion and ensure the usability of the generated magnetic resonance images.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] Acquire current magnetic resonance image data; the current magnetic resonance image data includes a magnetic resonance image of multiple contrasts of an upper slice, a magnetic resonance image of multiple contrasts of a current slice, and a magnetic resonance image of multiple contrasts of a lower slice;

[0007] Inputting the current magnetic resonance image data into a trained magnetic resonance image contrast conversion model to obtain a target contrast magnetic resonance image;

[0008] The process of determining the trained magnetic resonance image contrast conversion model specifically includes:

[0009] Constructing a magnetic resonance image sample data set, each sample of the magnetic resonance image sample data set includes sample magnetic resonance image data and a true contrast magnetic resonance image; the sample magnetic resonance image data includes a sample magnetic resonance image of multiple contrasts of an upper slice, a sample magnetic resonance image of multiple contrasts of a current slice, and a sample magnetic resonance image of multiple contrasts of a lower slice;

[0010] Constructing an initial magnetic resonance image contrast conversion model; the initial magnetic resonance image contrast conversion model includes a magnetic resonance image contrast conversion network and a discriminator network;

[0011] Input each sample of the magnetic resonance image sample data set into the magnetic resonance image contrast conversion network to obtain a predicted contrast magnetic resonance image, and input the predicted contrast magnetic resonance image, the true contrast magnetic resonance image and the sample magnetic resonance image data into a discriminator network to obtain a discriminator result;

[0012] A loss function is calculated according to the predicted contrast magnetic resonance image and the discriminator result to obtain a total loss value, and the initial magnetic resonance image contrast conversion model is updated according to the total loss value to obtain the trained magnetic resonance image contrast conversion model.

[0013] In a second aspect, the present application provides a computer device, comprising: 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 above-mentioned magnetic resonance image contrast conversion method.

[0014] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the above-mentioned magnetic resonance image contrast conversion method when executed by a processor.

[0015] In a fourth aspect, the present application provides a computer program product, including a computer program, which implements the above-mentioned magnetic resonance image contrast conversion method when executed by a processor.

[0016] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0017] The present application provides a method, device, medium and product for converting contrast of magnetic resonance images. By obtaining a target contrast magnetic resonance image based on magnetic resonance images of multiple contrasts, the model can combine the texture features of magnetic resonance images of multiple contrasts when performing contrast conversion, so that the converted target image has higher quality. At the same time, the input data includes magnetic resonance images of multiple adjacent slices, which helps the model understand the anatomical continuity of the image and ensures the spatial consistency and anatomical structure coherence of the generated target contrast image. Moreover, the generated image is evaluated by the discriminator network, and feedback is provided to the magnetic resonance image contrast conversion network for continuous adjustment and optimization to generate a higher quality and closer to the real target image. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 This is a diagram of an application environment of a magnetic resonance image contrast conversion method in an embodiment of the present application;

[0020] Figure 2 A schematic diagram of a flow chart of a magnetic resonance image contrast conversion method provided in one embodiment of the present application;

[0021] Figure 3 A schematic diagram of a process for determining a trained magnetic resonance image contrast conversion model provided in an embodiment of the present application;

[0022] Figure 4 A schematic diagram of functional modules of an initial magnetic resonance image contrast conversion model provided in an embodiment of the present application;

[0023] Figure 5 A schematic diagram of a swin-transformer network provided in an embodiment of the present application;

[0024] Figure 6 A schematic diagram of a hybrid convolutional swin-transformer network provided in an embodiment of the present application;

[0025] Figure 7 A schematic diagram of a real magnetic resonance image contrast conversion result provided by an embodiment of the present application;

[0026] Figure 8 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0028] Since the current magnetic resonance image contrast conversion method mainly focuses on single contrast and single slice data, it fails to fully consider the intrinsic structure and texture similarity of magnetic resonance images between multiple contrasts, as well as the anatomical continuity between consecutive slices. If multi-modal multi-slice information is taken into account at the same time, the quality of magnetic resonance image modality conversion will be further improved. Therefore, the present application provides a magnetic resonance image contrast conversion method.

[0029] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0030] The magnetic resonance image contrast conversion method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, or it can be integrated on the server 104, or it can be placed on the cloud or other servers. The terminal 102 can send the current magnetic resonance image data to the server 104. After the server 104 receives the current magnetic resonance image data, the server 104 inputs the current magnetic resonance image data into the trained magnetic resonance image contrast conversion model to obtain the target contrast magnetic resonance image. The server 104 can feed back the obtained target contrast magnetic resonance image to the terminal 102. In addition, in some embodiments, the magnetic resonance image contrast conversion method can also be implemented by the server 104 or the terminal 102 alone, such as the terminal 102 can directly perform conversion processing on the current magnetic resonance image data, or the server 104 can obtain the current magnetic resonance image data from the data storage system and perform conversion processing on the current magnetic resonance image data.

[0031] The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, tablet computers, and IoT devices, and the server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.

[0032] In an exemplary embodiment, Figure 2-Figure 3As shown, a method for converting contrast of magnetic resonance images is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used as an example to illustrate, including the following steps 201 to 202. Among them:

[0033] Step 201, obtaining current magnetic resonance image data, the current magnetic resonance image data is a magnetic resonance image with known contrast; the current magnetic resonance image data includes a magnetic resonance image with multiple contrasts of an upper slice, a magnetic resonance image with multiple contrasts of a current slice, and a magnetic resonance image with multiple contrasts of a lower slice, wherein the upper slice, the current slice, and the lower slice are three adjacent slices.

[0034] Step 202: input the current magnetic resonance image data into the trained magnetic resonance image contrast conversion model to obtain a target contrast magnetic resonance image.

[0035] Furthermore, the process of determining the trained magnetic resonance image contrast conversion model specifically includes:

[0036] Step 301, construct a magnetic resonance image sample data set, and divide 80% as a training data set, 5% as a verification data set, and 15% as a test data set; wherein each sample of the magnetic resonance image sample data set includes sample magnetic resonance image data and a true contrast magnetic resonance image, and the sample magnetic resonance image data includes sample magnetic resonance images of multiple contrasts of an upper slice, sample magnetic resonance images of multiple contrasts of a current slice, and sample magnetic resonance images of multiple contrasts of a lower slice, for example, the multiple contrasts include four contrasts of T1-weighted imaging, T2-weighted imaging, PD-weighted imaging, and Flair-weighted imaging.

[0037] Step 302, constructing an initial magnetic resonance image contrast conversion model; Figure 4 As shown, the initial MRI contrast conversion model includes a MRI contrast conversion network and a discriminator network connected in series.

[0038] Furthermore, the magnetic resonance image contrast conversion network includes a multi-contrast multi-slice fusion module and a contrast conversion module. The multi-contrast multi-slice fusion module includes three parallel-arranged fusion units, each of which includes an encoder and several swin-transformer networks connected in series in sequence. The contrast conversion module includes several hybrid convolutional swin-transformer networks connected in series in sequence and a decoder.

[0039] Furthermore, if Figure 5As shown, the swin-transformer network includes a block embedding layer, a tensor connection layer, a linear mapping layer, a window multi-head self-attention layer, a layer normalization layer, a multi-layer perception layer, a block inverse embedding layer and a convolution batch normalization layer; wherein the block embedding layer is used to divide the input magnetic resonance images of multiple contrasts into multiple non-overlapping small blocks, the tensor connection layer is used to merge the tensor information of the magnetic resonance images of multiple contrasts of the upper slice, the magnetic resonance images of multiple contrasts of the current slice and the magnetic resonance images of multiple contrasts of the lower slice, the linear mapping layer is used to perform linear transformation on the output data of the tensor connection layer, the window multi-head self-attention layer is used to perform self-attention calculation on the output data of the linear mapping layer in each window, the layer normalization layer is used to normalize the output data of the window multi-head self-attention layer, the multi-layer perception layer is a feedforward neural network composed of multiple fully connected layers, and is used to further extract features from the output data of the layer normalization layer to extract higher-level feature representations, and the block inverse embedding layer is used to merge the multiple non-overlapping small blocks after the above processing into an image so that they can be used for convolution operations later.

[0040] Furthermore, if Figure 6 As shown in the figure, the hybrid convolutional swin-transformer network includes two parts: residual convolutional network and residual swin-transformer network; the residual convolutional network includes multiple residual connected convolutional layers to extract local features of the image, and the residual connection is realized by skipping multiple layers of input features and adding them to the final output features. The residual swin-transformer network includes a block embedding layer, a position encoding layer, a linear mapping layer, a sliding window multi-head self-attention layer, a layer normalization layer, a multi-layer perception layer, a block de-embedding layer, and a convolutional batch normalization layer.

[0041] Furthermore, the discriminator network is a conditional discriminator, the core component of which is a convolutional unit, and is organized in a cascade manner through multiple convolutional layers. The input of this discriminator network includes two parts of data: one is the input contrast magnetic resonance image, and the other is the true contrast magnetic resonance image (or the contrast conversion result generated by the network, that is, the predicted contrast magnetic resonance image), and its output is a value between 0 and 1. When the input of the discriminator network is the predicted contrast magnetic resonance image, the goal of the generator (that is, the magnetic resonance image contrast conversion network) is to deceive the discriminator (discriminator network) so that it thinks that the generated data is real data; when the input is the true contrast magnetic resonance image, the goal of the generator is to deceive the discriminator so that it thinks that the true data is false data.

[0042] Step 303, input each sample of the magnetic resonance image sample data set into the magnetic resonance image contrast conversion network to obtain a predicted contrast magnetic resonance image, and input the predicted contrast magnetic resonance image, the true contrast magnetic resonance image and the sample magnetic resonance image data into the discriminator network to obtain a discriminator result, specifically including:

[0043] Step 3031: slice the sample magnetic resonance image X of multiple contrasts up , sample magnetic resonance image X of multiple contrasts of the current slice cur Magnetic resonance images of samples with multiple contrasts and slices below X down Input them into any fusion unit respectively, and get the initial features of the slices above through the encoder. Initial features of the current slice and the initial features of the slice below And through several swin-transformer networks, the fusion features of the upper slices are obtained respectively The fusion features of the current slice And the fusion features of the slice below As shown below:

[0044]

[0045]

[0046]

[0047]

[0048] in, represents a sample MRI image with contrast 1 in the upper slice, The sample MRI image with contrast 2 in the upper slice is shown. represents the sample magnetic resonance image of the upper slice with contrast n; represents the sample MRI image with contrast 1 in the current slice, represents the sample MRI image of contrast 2 in the current slice, represents the sample magnetic resonance image of contrast n in the current slice; represents a sample MRI image with contrast 1 in the lower slice, A sample MRI image showing contrast 2 in the lower slice, represents a sample magnetic resonance image of contrast n in the slice below; M represents the swin-transformer network, and N represents the number of swin-transformer networks.

[0049] Step 3032, selecting the fusion feature of the current slice and determining it as the overall fusion feature.

[0050] The overall fusion feature is input into the contrast conversion module, and the final fusion feature is obtained through several hybrid convolutional swin-transformer networks. As shown below:

[0051]

[0052] Among them, H represents the hybrid convolutional swin-transformer network, and Q represents the number of hybrid convolutional swin-transformer networks.

[0053] Step 3033: input the final fusion feature into the decoder to obtain the predicted contrast magnetic resonance image X trans , as shown below:

[0054]

[0055] Among them, F dec For the decoder.

[0056] Step 3034: predict the contrast magnetic resonance image X trans , true contrast MRI image X tar and sample MRI data X source Input to the discriminator network and get the discriminator result,

[0057] Step 304, calculating the loss function according to the predicted contrast magnetic resonance image and the discriminator result to obtain a total loss value, and updating the initial magnetic resonance image contrast conversion model according to the total loss value to obtain a trained magnetic resonance image contrast conversion model.

[0058] Further loss functions include pixel-level consistency loss and adversarial loss. The pixel-level consistency loss is measured using the L1 norm, and the adversarial loss is measured using a conditional discriminator, which significantly improves the image conversion quality. The formula is:

[0059] L syn =L pixel +L adv ;

[0060] Among them, L syn Represents the total loss value, L pixel Represents the pixel-level consistency loss value, L adv Represents the adversarial level loss value.

[0061] The pixel-level consistency loss is specifically:

[0062]

[0063] Where i represents the image pixel, I represents the number of image pixels, E represents the expectation, and F gen represents the magnetic resonance image contrast conversion network, X source represents the sample magnetic resonance image data, X tar represents the true contrast MRI image, F gen (X source ) is to convert the sample magnetic resonance image data to generate a predicted contrast magnetic resonance image, and ||·||1 is to calculate the sum of the absolute differences in pixel level between the predicted contrast magnetic resonance image and the true contrast magnetic resonance image.

[0064] The adversarial level loss is:

[0065] L adv =-E[F dis ([X source ,X tar ]) 2 ]-E[(F dis ([X source ,X trans ])-1) 2 ];

[0066] Among them, F dis represents the discriminator network, E represents the expectation, X source represents the sample magnetic resonance image data, X tar represents the true contrast MRI image, X trans represents the predicted contrast MRI image, F dis ([X source ,X tar ]) is the result of distinguishing the sample magnetic resonance image data from the predicted contrast magnetic resonance image, F dis ([X source ,X trans ]) is the result of distinguishing the sample magnetic resonance image data from the true contrast magnetic resonance image.

[0067] according to Figure 7According to the real magnetic resonance image contrast conversion results recorded in , it can be known that the implementation of the above-mentioned magnetic resonance image contrast conversion method can effectively improve the quality of the magnetic resonance image after contrast conversion. The present application makes full use of multi-contrast and multi-slice information, adopts multiple contrasts and three adjacent slices as input data, and the number of input contrasts is not limited, and makes full use of the complementarity of tissue structure information between different contrasts, as well as the gradual change of adjacent slices in tissue structure, and based on this, designs a three-branch fusion module based on the swin-transformer network to fully fuse multi-contrast and multi-slice information. Moreover, in order to make full use of local information and global information, a contrast conversion module based on a hybrid convolutional swin-transformer network is designed, which makes full use of the advantages of convolutional neural networks in local feature extraction and the advantages of swin-transformer in global feature extraction to improve image conversion quality.

[0068] The magnetic resonance image contrast conversion method studied in this application has a wide range of applications. It can not only be used for single-contrast to single-contrast conversion, such as when the magnetic resonance device collects T2 contrast, it can realize tasks such as T2→T1, T2→PD and T2→Flair, but also can be used for multi-contrast to single-contrast conversion, such as when the magnetic resonance device collects T1 and T2 contrast, it can realize tasks such as {T1, T2}→PD, {T1, T2}→Flair, etc.

[0069] The present application also provides an application scenario, which applies the above-mentioned magnetic resonance image contrast conversion method. Specifically: the magnetic resonance image contrast conversion method provided in this embodiment can be applied in the scenario of generating a new target contrast magnetic resonance image. The scenario includes a target contrast magnetic resonance image generation link and a target contrast magnetic resonance image display link; the target contrast magnetic resonance image generation link is used to obtain a new target contrast magnetic resonance image based on a known plurality of slices and a plurality of contrast magnetic resonance images through a magnetic resonance image contrast conversion model; the target contrast magnetic resonance image display link is used to display the generated target contrast magnetic resonance image for reference and diagnosis by medical staff. The magnetic resonance image contrast conversion method provided in this embodiment belongs to the target contrast magnetic resonance image generation link.

[0070] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 8As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used for magnetic resonance image data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a magnetic resonance image contrast conversion method is implemented.

[0071] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0072] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0073] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0074] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0075] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0076] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0077] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0078] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for converting contrast of a magnetic resonance image, characterized in that: The magnetic resonance image contrast conversion method comprises: Acquire current magnetic resonance image data; the current magnetic resonance image data includes a magnetic resonance image of multiple contrasts of an upper slice, a magnetic resonance image of multiple contrasts of a current slice, and a magnetic resonance image of multiple contrasts of a lower slice; Inputting the current magnetic resonance image data into a trained magnetic resonance image contrast conversion model to obtain a target contrast magnetic resonance image; The process of determining the trained magnetic resonance image contrast conversion model specifically includes: Constructing a magnetic resonance image sample data set, each sample of the magnetic resonance image sample data set includes sample magnetic resonance image data and a true contrast magnetic resonance image; the sample magnetic resonance image data includes a sample magnetic resonance image of multiple contrasts of an upper slice, a sample magnetic resonance image of multiple contrasts of a current slice, and a sample magnetic resonance image of multiple contrasts of a lower slice; Constructing an initial magnetic resonance image contrast conversion model; the initial magnetic resonance image contrast conversion model includes a magnetic resonance image contrast conversion network and a discriminator network; Input each sample of the magnetic resonance image sample data set into the magnetic resonance image contrast conversion network to obtain a predicted contrast magnetic resonance image, and input the predicted contrast magnetic resonance image, the true contrast magnetic resonance image and the sample magnetic resonance image data into a discriminator network to obtain a discriminator result; A loss function is calculated according to the predicted contrast magnetic resonance image and the discriminator result to obtain a total loss value, and the initial magnetic resonance image contrast conversion model is updated according to the total loss value to obtain the trained magnetic resonance image contrast conversion model.

2. The magnetic resonance image contrast conversion method according to claim 1, characterized in that: The magnetic resonance image contrast conversion network includes a multi-contrast multi-slice fusion module and a contrast conversion module connected in series; The multi-contrast multi-slice fusion module includes three parallel fusion units; each fusion unit includes an encoder and a plurality of swin-transformer networks connected in series in sequence; The contrast conversion module includes a plurality of hybrid convolutional swin-transformer networks connected in series and a decoder.

3. The magnetic resonance image contrast conversion method according to claim 2, characterized in that: The swin-transformer network includes a block embedding layer, a tensor connection layer, a linear mapping layer, a window multi-head self-attention layer, a layer normalization layer, a multi-layer perception layer, a block de-embedding layer and a convolutional batch normalization layer; wherein the tensor connection layer is used to merge the tensor information of the magnetic resonance images of multiple contrasts of the upper slice, the magnetic resonance images of multiple contrasts of the current slice and the magnetic resonance images of multiple contrasts of the lower slice.

4. The magnetic resonance image contrast conversion method according to claim 2, characterized in that: The hybrid convolutional swin-transformer network includes a residual convolutional network and a residual swin-transformer network; wherein the residual convolutional network includes a plurality of residually connected convolutional layers, and the residual swin-transformer network includes a block embedding layer, a position encoding layer, a linear mapping layer, a sliding window multi-head self-attention layer, a layer normalization layer, a multi-layer perception layer, a block de-embedding layer and a convolutional batch normalization layer.

5. The magnetic resonance image contrast conversion method according to claim 2, characterized in that: Inputting each sample of the magnetic resonance image sample data set into the magnetic resonance image contrast conversion network to obtain a predicted contrast magnetic resonance image specifically includes: The sample magnetic resonance images of the upper slice with multiple contrasts, the sample magnetic resonance images of the current slice with multiple contrasts, and the sample magnetic resonance images of the lower slice with multiple contrasts are respectively input into any one of the fusion units, and the initial features of the upper slice, the initial features of the current slice, and the initial features of the lower slice are respectively obtained through the encoder, and the fusion features of the upper slice, the fusion features of the current slice, and the fusion features of the lower slice are respectively obtained through a plurality of swin-transformer networks; Selecting the fusion feature of the current slice to determine as the overall fusion feature; The overall fusion feature is input into the contrast conversion module, and the final fusion feature is obtained through several hybrid convolutional swin-transformer networks; The final fused features are input into a decoder to obtain the predicted contrast magnetic resonance image.

6. The magnetic resonance image contrast conversion method according to claim 1, characterized in that: The discriminator network is a conditional discriminator, which includes multiple convolutional layers in a cascade manner.

7. The magnetic resonance image contrast conversion method according to claim 1, characterized in that: The loss function includes pixel-level consistency loss and adversarial level loss; The pixel-level consistency loss is specifically: Where i represents the image pixel, I represents the number of image pixels, E represents the expectation, and F gen represents the magnetic resonance image contrast conversion network, X source represents the sample magnetic resonance image data, X tar represents the true contrast MRI image, F gen (X source ) is to convert the sample magnetic resonance image data to generate a predicted contrast magnetic resonance image, ||·||1 is to calculate the sum of the absolute differences in pixel level between the predicted contrast magnetic resonance image and the true contrast magnetic resonance image; The adversarial level loss is specifically: L adv =-E[F dis ([X source ,X tar ]) 2 ]-E[(F dis ([X source ,X trans ])-1) 2 ]; Among them, F dis represents the discriminator network, E represents the expectation, X source represents the sample magnetic resonance image data, X tar represents the true contrast MRI image, X trans represents the predicted contrast MRI image, F dis ([X source ,X tar ]) is the result of distinguishing the sample magnetic resonance image data from the predicted contrast magnetic resonance image, F dis ([X source ,X trans ]) is the result of distinguishing the sample magnetic resonance image data from the true contrast magnetic resonance image.

8. A computer device comprising: 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 image contrast conversion method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the magnetic resonance image contrast conversion method according to any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the magnetic resonance image contrast conversion method according to any one of claims 1 to 7 is implemented.

Images