Brain magnetic resonance image super-resolution reconstruction method and system based on implicit representation model

Abstract

The invention provides a brain magnetic resonance image super-resolution reconstruction method and system based on a hidden representation model, and the method comprises the steps: obtaining anisotropic brain magnetic resonance images with different layer thicknesses, and dividing the anisotropic brain magnetic resonance images into training data and test data; preprocessing the brain magnetic resonance image; processing the preprocessed training data to generate a training sample; training the super-resolution reconstruction model based on the hidden representation model by adopting the training sample; and inputting the preprocessed test data into the trained model, stacking to obtain a three-dimensional image, and averaging to finally obtain an isotropic brain magnetic resonance image. According to the method, super-resolution reconstruction of anisotropic brain magnetic resonance images with different layer thicknesses is carried out by using a deep learning super-resolution reconstruction method, a learnable coordinate coding module based on a hidden representation model is designed, the correlation between image features and coordinate information is fully utilized, a mixed loss function is designed, and the super-resolution reconstruction of the anisotropic brain magnetic resonance images with different layer thicknesses is realized. And the accuracy of super-resolution reconstruction of anisotropic brain magnetic resonance images with different layer thicknesses is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a method and system for super-resolution reconstruction of brain magnetic resonance images based on an implicit representation model. Background Art

[0002] Magnetic resonance imaging is a non-invasive, radiation-free medical imaging technique that is widely used in disease detection, diagnosis, treatment progress, and prognosis monitoring. However, due to equipment limitations and imaging conditions, clinical practice usually uses a two-dimensional scanning protocol to obtain brain MR images. The brain MR images obtained in this way have the characteristics of anisotropic voxel resolution, which is specifically manifested in that the resolution of the inter-layer (i.e., the direction of the scanning layer selection) image is relatively low, and the resolution of the intra-layer (i.e., the direction perpendicular to the scanning layer selection) image is relatively high. Due to the large layer thickness of these anisotropic brain MR images, the scanning layer selection direction will present unnatural image textures due to aliasing. In addition, the scanning parameters of different hospitals may also vary, so that even the same brain MR sequence will obtain anisotropic brain MR data with different layer thicknesses. These anisotropic brain MR data are prone to partial volume effects, which brings many difficulties to subsequent image processing, analysis, and disease diagnosis. Therefore, improving the quality and resolution of MR images is of great clinical significance.

[0003] As a post-processing work, super-resolution (SR) technology has been widely used to improve the quality of magnetic resonance images. Traditional super-resolution methods include two categories: one is based on interpolation methods, such as nearest neighbor interpolation and bicubic interpolation; the other is based on regularization methods, which effectively constrain the range of solutions by integrating prior knowledge or introducing regularization terms to improve the quality of reconstructed images, such as non-local means and low-rank total variation. In recent years, with the development of deep learning technology, especially convolutional neural networks, which directly learn the mapping from low-resolution (LR) images to high-resolution (HR) images with their excellent high-level abstract representation learning ability, although significant image restoration results have been achieved, these deep learning-based methods usually focus on the same layer thickness fixed upsampling factor, with limited flexibility, which limits the application of these methods in clinical scenarios.

[0004] Among the related technologies, there are the Chinese invention patent with application number CN202410001400.8, whose invention name is a multi-contrast magnetic resonance image super-resolution method based on feature fusion Transformer; the Chinese invention patent with application number CN202210049071.5, whose invention name is a brain magnetic resonance image super-resolution reconstruction method; and the Chinese invention patent with application number CN202311439600.3, whose invention name is a magnetic resonance image super-resolution reconstruction method based on multi-scale attention.

[0005] The above methods require paired high-resolution and low-resolution magnetic resonance images and adopt supervised training strategies for training. However, in actual clinical applications, paired high-resolution and low-resolution magnetic resonance images are often lacking. Existing super-resolution reconstruction methods based on self-supervised learning use the information of anisotropic magnetic resonance images themselves to generate paired training samples, generate low-resolution images from high-resolution slices within the layer for training, and then use the trained model for the other two inter-layer slices. However, such methods usually perform personalized model training for each anisotropic magnetic resonance image, which has limited flexibility. The method that combines implicit neural representation with self-supervised learning methods realizes super-resolution reconstruction of anisotropic magnetic resonance images of different layer thicknesses by fusing image features with coordinate information. However, this method uses a direct feature splicing method when fusing features, which makes it difficult to fully utilize the correlation between the two. Summary of the invention

[0006] In order to achieve the above-mentioned purpose and other advantages of the present invention, the first purpose of the present invention is to provide a brain magnetic resonance image super-resolution reconstruction method based on an implicit representation model, comprising the following steps:

[0007] Acquire anisotropic brain magnetic resonance images of different slice thicknesses and divide them into training data and test data;

[0008] Preprocessing the anisotropic brain magnetic resonance image;

[0009] Processing the preprocessed training data to generate training samples;

[0010] Using the training samples to train a super-resolution reconstruction model based on a latent representation model;

[0011] Input the preprocessed test data into a trained model, output a reconstructed image slice image, and stack them to obtain a three-dimensional image;

[0012] The obtained three-dimensional images are then averaged to finally obtain an isotropic brain magnetic resonance image.

[0013] Furthermore, the step of preprocessing the anisotropic brain magnetic resonance image includes:

[0014] performing bias field correction on the anisotropic brain magnetic resonance image;

[0015] The pixel values ​​of the anisotropic brain magnetic resonance image are subjected to maximum and minimum normalization processing.

[0016] Furthermore, the step of processing the preprocessed training data to generate training samples includes:

[0017] Extracting high-resolution image blocks according to preset sizes from high-resolution intra-layer slices;

[0018] According to the ratio of the inter-layer and intra-layer voxel resolutions, Gaussian blurring and one-dimensional downsampling are performed on the high-resolution image block, and the downsampling scale is the ratio;

[0019] The downsampled image blocks are upsampled to obtain low-resolution image blocks of the same size as the high-resolution image blocks, and constitute a set of training samples.

[0020] Furthermore, the super-resolution reconstruction model based on the implicit representation model includes an image feature extraction module, a learnable coordinate encoding module and a decoding module; wherein,

[0021] The image feature extraction module is used to extract image features from image blocks;

[0022] The learnable coordinate encoding module is used to extract coordinate encoding from image block coordinates;

[0023] The decoding module is used to decode the fusion feature of the image feature and the coordinate code.

[0024] Furthermore, the image feature extraction module adopts an EDSR-based network structure, which includes a convolutional layer, a residual block and a jump connection. The residual block sequentially includes a convolution, a linear rectifier unit, a convolution and a constant scaling layer. The jump connection connects the feature map output by the first convolutional layer with the output feature map of the penultimate convolutional layer, and sends the connected features to the last convolutional layer.

[0025] Furthermore, the learnable coordinate encoding module includes multiple fully connected layers, whose input feature dimensions correspond to the coordinates of each pixel in the image. The first fully connected layer maps the input features to a first preset dimensional space and performs a nonlinear transformation through an activation function. The middle fully connected layer further expands the multi-dimensional features of the previous fully connected layer to more dimensions and is also followed by an activation function. The penultimate fully connected layer maintains the multi-dimensional features of the previous fully connected layer and is also followed by an activation function. The last connected layer maps the multi-dimensional features to the final output feature dimensions.

[0026] Furthermore, the decoding module includes multiple fully connected layers, each of which is followed by an activation function. A residual connection is also introduced in the decoding module to add the input information to the output of the intermediate activation layer.

[0027] Furthermore, the loss function for training the super-resolution reconstruction model based on the implicit representation model using the training samples adopts a hybrid loss function that integrates content loss and brain tissue edge perception loss, and the formula is defined as follows:

[0028]

[0029] Among them, y represents the real image, Represents the prediction result of super-resolution reconstruction, L c Indicates content loss, L eg represents the edge perception loss of brain tissue, and weights λ1 and λ2 are used to balance different loss function terms; L c , L eg The definition is as follows:

[0030]

[0031]

[0032] Among them, C h and C w Represents the horizontal and vertical edge detection Prewitt operators, represents the convolution operation, M y and Represent the real graph of y after the Prewitt operator and Gradient magnitude map of the prediction image.

[0033] The second object of the present invention is to provide a brain magnetic resonance image super-resolution reconstruction system based on an implicit representation model, which implements the above method and includes an image acquisition module, an image preprocessing module, a training sample generation module, a model training module, a model testing module, and an image processing module; wherein,

[0034] The image acquisition module is used to acquire anisotropic brain magnetic resonance images of different layer thicknesses, which are divided into training data and test data;

[0035] The image preprocessing module is used to preprocess the anisotropic brain magnetic resonance image;

[0036] The training sample generating module is used to process the pre-processed training data to generate training samples;

[0037] The model training module is used to train a super-resolution reconstruction model based on a latent representation model using the training samples;

[0038] The model testing module is used to input the preprocessed test data into the trained model, output the reconstructed image slice images, and stack them to obtain a three-dimensional image;

[0039] The image processing module is used to average the obtained three-dimensional images to finally obtain an isotropic brain magnetic resonance image.

[0040] A third object of the present invention is to provide a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0041] A fourth object of the present invention is to provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] The present invention uses a deep learning super-resolution reconstruction method to perform super-resolution reconstruction of anisotropic brain magnetic resonance images with different layer thicknesses. By designing a learnable coordinate encoding module based on an implicit representation model, the correlation between image features and coordinate information is fully utilized. At the same time, a hybrid loss function of fusion loss and brain tissue edge perception loss is designed, so that the gray-white matter edge and brain tissue texture information are paid attention to during network training, thereby improving the accuracy of super-resolution reconstruction of anisotropic brain magnetic resonance images with different layer thicknesses.

[0044] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention in conjunction with the accompanying drawings. The specific implementation of the present invention is given in detail by the following embodiments and their accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0046] Figure 1 This is a flow chart of the method for super-resolution reconstruction of brain magnetic resonance images based on an implicit representation model of Example 1;

[0047] Figure 2 This is a flowchart of anisotropic brain magnetic resonance image preprocessing in Example 1;

[0048] Figure 3 The flowchart of generating training samples in Example 1;

[0049] Figure 4 This is a schematic diagram of the overall network framework of Example 1;

[0050] Figure 5 This is a schematic diagram of an image feature extraction module in Example 1;

[0051] Figure 6 This is a schematic diagram of a learnable coordinate encoding module of Example 1;

[0052] Figure 7 This is a schematic diagram of a decoding module of Example 1;

[0053] Figure 8 This is a flow chart of the training phase of Example 1;

[0054] Fig. 9 This is a flow chart of the testing phase of Example 1;

[0055] Fig.10 Schematic diagram of visualization results when the super-resolution scales are 2 and 4 in Example 1;

[0056] Fig.11 Schematic diagram of a brain magnetic resonance image super-resolution reconstruction system based on an implicit representation model according to Example 2;

[0057] Fig.12 This is a schematic diagram of a computer device in Example 3;

[0058] Fig.13 This is a schematic diagram of a computer-readable storage medium of Example 4. DETAILED DESCRIPTION

[0059] The present invention is further described below in conjunction with the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. It should be noted that, under the premise of no conflict, the embodiments or technical features described below can be arbitrarily combined to form a new embodiment.

[0060] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making any creative work shall fall within the scope of protection of the present invention.

[0061] The figure numbers in this application are only used to distinguish the various steps in the scheme, and are not used to limit the execution order of the various steps. The specific execution order is subject to the description in the specification.

[0062] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0063] Example 1

[0064] A super-resolution reconstruction method for brain magnetic resonance images based on implicit representation model, such as Figure 1 As shown, the following steps are included:

[0065] S1. Obtain anisotropic brain magnetic resonance images of different slice thicknesses and divide them into training data and test data;

[0066] S2, preprocessing the anisotropic brain magnetic resonance image;

[0067] In some embodiments, Figure 2 As shown, the step of preprocessing the anisotropic brain magnetic resonance image includes:

[0068] S21, performing bias field correction on the anisotropic brain magnetic resonance image to improve the grayscale uniformity of the image;

[0069] S22, applying maximum and minimum normalization processing to the pixel values ​​of the anisotropic brain magnetic resonance image, so as to map the data to between 0 and 1.

[0070] S3, processing the preprocessed training data to generate training samples;

[0071] In some embodiments, Figure 3 As shown, the step of processing the preprocessed training data to generate training samples includes:

[0072] S31 . Extracting a high-resolution image block according to a preset size from a high-resolution intra-layer slice; for example, extracting a high-resolution image block with a size of 48*48 from a high-resolution intra-layer slice.

[0073] S32, performing Gaussian blur and one-dimensional downsampling on the high-resolution image block according to the ratio of the inter-layer and intra-layer voxel resolutions, with the downsampling scale being the ratio;

[0074] S33, performing an upsampling operation on the downsampled image block to obtain a low-resolution image block with the same size as the high-resolution image block, and forming a group of training samples.

[0075] S4, using the training samples to train a super-resolution reconstruction model based on an implicit representation model;

[0076] In some embodiments, Figure 4 , Figure 8 As shown, the super-resolution reconstruction model based on the hidden representation model includes an image feature extraction module, a learnable coordinate encoding module and a decoding module; wherein,

[0077] The image feature extraction module is used to extract image features from image blocks;

[0078] The learnable coordinate encoding module is used to extract coordinate encoding from image block coordinates;

[0079] The decoding module is used to decode the fusion feature of the image feature and the coordinate code.

[0080] like Figure 5 As shown, the image feature extraction module adopts a network structure based on EDSR, which includes a convolution layer, a residual block and a jump connection. For example, the step size is 1*1, the padding is 1*1, and the convolution kernel size is 3*3. The residual block contains convolution, linear rectifier unit, convolution and constant scaling layer in sequence. For example, a total of 16 residual blocks are used, and each residual block contains a 3*3 convolution, a linear rectifier unit (ReLU), a 3*3 convolution and a constant scaling layer with a residual scaling factor of 0.1 in sequence. The jump connection connects the feature map output by the first convolution layer with the output feature map of the second to last convolution layer, and finally sends the connected features to the last convolution layer, for example, the output channel is 128.

[0081] The learnable coordinate encoding module includes multiple fully connected layers, whose input feature dimensions correspond to the coordinates of each pixel in the image. The first fully connected layer maps the input features to a first preset dimensional space and performs a nonlinear transformation through an activation function. The middle fully connected layer further expands the multi-dimensional features of the previous fully connected layer to more dimensions and is also followed by an activation function. The penultimate fully connected layer maintains the multi-dimensional features of the previous fully connected layer and is also followed by an activation function. The last connected layer maps the multi-dimensional features to the final output feature dimensions.

[0082] like Figure 6 As shown in the figure, the learnable coordinate encoding module specifically includes 5 fully connected layers (FC Layer). The input feature dimension of the network is 2, corresponding to the two-dimensional coordinates of each pixel in the image. Subsequently, the first FC layer maps the input features to a 64-dimensional space and performs a nonlinear transformation through a ReLU activation function. Next, the second FC layer further expands the 64-dimensional features to 128 dimensions, and is also followed by a ReLU activation function. The third FC layer continues this process, increasing the feature dimension to 256 dimensions and maintaining the use of the ReLU activation function. The fourth FC layer maintains a 256-dimensional feature space and applies the ReLU activation function again to enhance the nonlinear expression ability of the network. Finally, the fifth FC layer maps the 256-dimensional features to a final output feature dimension of 128.

[0083] The decoding module includes multiple fully connected layers, each of which is followed by an activation function. A residual connection is also introduced in the decoding module to add the input information to the output of the intermediate activation layer.

[0084] like Figure 7As shown in the figure, the decoding module consists of eight FC layers. In order to improve the nonlinear expression and learning ability of the network, each fully connected layer is followed by a ReLU activation function. In addition, a residual connection is introduced in the network to add the input information to the output of the fourth ReLU activation layer. The input and output feature dimensions of the first 7 FC layers are 256 and 256, and the input and output feature dimensions of the last FC layer are 256 and 1.

[0085] In some embodiments, the model is initialized and then trained using data from training samples. The loss function in the training process uses a hybrid loss function that integrates content loss and brain tissue edge perception loss, and the formula is defined as follows:

[0086]

[0087] Among them, y represents the real image, Represents the prediction result of super-resolution reconstruction, L c Indicates content loss, L eg represents the edge perception loss of brain tissue, and weights λ1 and λ2 are used to balance different loss function terms; L c , L eg The definition is as follows:

[0088]

[0089] Among them, C h and C w Represents the horizontal and vertical edge detection Prewitt operators, represents the convolution operation, M y and Represent the real graph of y after the Prewitt operator and Gradient magnitude map of the prediction image.

[0090] S5, inputting the preprocessed test data into the trained model, outputting the reconstructed image slice images, and stacking them to obtain a three-dimensional image;

[0091] S6. The obtained three-dimensional images are averaged to finally obtain an isotropic brain magnetic resonance image.

[0092] like Fig. 9 As shown, when the model is tested, the inter-layer slices are input into the above-mentioned trained model, and are stacked to obtain three-dimensional images, and then the three-dimensional images obtained are averaged to finally obtain an isotropic brain magnetic resonance image.

[0093] The visualization results of super-resolution reconstructed images and original images using different methods are shown in the figure below. Fig.10 shown.

[0094] The super-resolution reconstruction performance indicators using different methods are shown in Table 1.

[0095] The results show that the peak signal-to-noise ratio (PSNR) and average structural similarity (SSIM) of the brain magnetic resonance image super-resolution reconstruction method based on the implicit representation model provided by the present invention are optimal.

[0096] Table 1 Super-resolution reconstruction performance indicators

[0097] method PSNR SSIM Traditional interpolation method 34.88 0.9537 Method 1 35.26 0.9591 Method 2 39.80 0.9771 Method 3 40.17 0.9753 The method proposed by the present invention 40.57 0.9785

[0098] This embodiment proposes a super-resolution reconstruction network based on an implicit representation model, designs a learnable coordinate encoding module, and converts coordinate information into learned coordinate encoding information through a network implicit representation learning mechanism, so that the fusion process pays more attention to the correlation between coordinate information and image features, thereby improving the performance of local implicit image functions; this embodiment designs a hybrid loss function that integrates content loss and brain tissue edge perception loss, which can take into account both the overall and local characteristics of the image during the training process, thereby generating super-resolution images with clearer textures and higher quality.

[0099] Example 2

[0100] A brain magnetic resonance image super-resolution reconstruction system 700 based on an implicit representation model implements the above method. For a detailed description of the method, reference can be made to the corresponding description in the above method embodiment, which will not be repeated here. Fig.11 As shown, the system includes an image acquisition module 710, an image preprocessing module 720, a training sample generation module 730, a model training module 740, a model testing module 750, and an image processing module 760; wherein,

[0101] The image acquisition module is used to acquire anisotropic brain magnetic resonance images of different layer thicknesses, which are divided into training data and test data;

[0102] The image preprocessing module is used to preprocess the anisotropic brain magnetic resonance image;

[0103] The training sample generating module is used to process the pre-processed training data to generate training samples;

[0104] The model training module is used to train a super-resolution reconstruction model based on a latent representation model using the training samples;

[0105] The model testing module is used to input the preprocessed test data into the trained model, output the reconstructed image slice images, and stack them to obtain a three-dimensional image;

[0106] The image processing module is used to average the obtained three-dimensional images to finally obtain an isotropic brain magnetic resonance image.

[0107] Based on the technical solution of the above embodiment, optionally, the step of preprocessing the anisotropic brain magnetic resonance image includes:

[0108] performing bias field correction on the anisotropic brain magnetic resonance image;

[0109] The pixel values ​​of the anisotropic brain magnetic resonance image are subjected to maximum and minimum normalization processing.

[0110] Based on the technical solution of the above embodiment, optionally, the step of processing the preprocessed training data to generate training samples includes:

[0111] Extracting high-resolution image blocks according to preset sizes from high-resolution intra-layer slices;

[0112] According to the ratio of the inter-layer and intra-layer voxel resolutions, Gaussian blurring and one-dimensional downsampling are performed on the high-resolution image block, and the downsampling scale is the ratio;

[0113] The downsampled image blocks are upsampled to obtain low-resolution image blocks of the same size as the high-resolution image blocks, and constitute a set of training samples.

[0114] Based on the technical solution of the above embodiment, optionally, the super-resolution reconstruction model based on the implicit representation model includes an image feature extraction module, a learnable coordinate encoding module and a decoding module; wherein,

[0115] The image feature extraction module is used to extract image features from image blocks;

[0116] The learnable coordinate encoding module is used to extract coordinate encoding from image block coordinates;

[0117] The decoding module is used to decode the fusion feature of the image feature and the coordinate code.

[0118] Based on the technical solution of the above embodiment, optionally, the image feature extraction module adopts a network structure based on EDSR, which includes a convolution layer, a residual block and a jump connection. The residual block sequentially includes a convolution, a linear rectification unit, a convolution and a constant scaling layer. The jump connection connects the feature map output by the first convolution layer with the output feature map of the penultimate convolution layer, and sends the connected features to the last convolution layer.

[0119] Based on the technical solutions of the above embodiments, optionally, the learnable coordinate encoding module includes multiple fully connected layers, whose input feature dimensions correspond to the coordinates of each pixel in the image. The first fully connected layer maps the input features to a first preset dimensional space and performs a nonlinear transformation through an activation function. The middle fully connected layer further expands the multi-dimensional features of the previous fully connected layer to more dimensions and is also followed by an activation function. The penultimate fully connected layer maintains the multi-dimensional features of the previous fully connected layer and is also followed by an activation function. The last connected layer maps the multi-dimensional features to the final output feature dimensions.

[0120] Based on the technical solution of the above embodiment, optionally, the decoding module includes multiple fully connected layers, each fully connected layer is followed by an activation function, and a residual connection is also introduced in the decoding module to add the input information to the output of the intermediate activation layer.

[0121] Based on the technical solution of the above embodiment, optionally, the loss function for training the super-resolution reconstruction model based on the implicit representation model using the training samples adopts a hybrid loss function that integrates content loss and brain tissue edge perception loss, and the formula is defined as follows:

[0122]

[0123] Among them, y represents the real image, Represents the prediction result of super-resolution reconstruction, L c Indicates content loss, L eg represents the edge perception loss of brain tissue, and weights λ1 and λ2 are used to balance different loss function terms; L c , L eg The definition is as follows:

[0124]

[0125] Among them, C h and C w Represents the horizontal and vertical edge detection Prewitt operators, represents the convolution operation, M y and Represent the real graph of y after the Prewitt operator and Gradient magnitude map of the prediction image.

[0126] This embodiment proposes a super-resolution reconstruction network based on an implicit representation model, designs a learnable coordinate encoding module, and converts coordinate information into learned coordinate encoding information through a network implicit representation learning mechanism, so that the fusion process pays more attention to the correlation between coordinate information and image features, thereby improving the performance of local implicit image functions; this embodiment designs a hybrid loss function that integrates content loss and brain tissue edge perception loss, which can take into account both the overall and local characteristics of the image during the training process, thereby generating super-resolution images with clearer textures and higher quality.

[0127] Example 3

[0128] A computer device 800, such as Fig.12 As shown, it includes a memory 810, a processor 820, and a computer program 830 stored in the memory and executable on the processor. When the processor executes the computer program, the steps of a method for super-resolution reconstruction of brain magnetic resonance images based on an implicit representation model are implemented. For a detailed description of the method, reference may be made to the corresponding description in the above method embodiment, which will not be repeated here.

[0129] Example 4

[0130] A computer readable storage medium such as Fig.13 As shown, a computer program is stored thereon, and when the computer program is executed by the processor, the steps of a method for super-resolution reconstruction of brain magnetic resonance images based on an implicit representation model are implemented. For a detailed description of the method, reference may be made to the corresponding description in the above method embodiment, and no further description is given here.

[0131] Example 5

[0132] A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the steps of a method for super-resolution reconstruction of brain magnetic resonance images based on an implicit representation model are implemented. For a detailed description of the method, reference may be made to the corresponding description in the above method embodiment, which will not be repeated here.

[0133] The number of devices and processing scales described here are used to simplify the description of the present invention. Applications, modifications and variations of the present invention will be obvious to those skilled in the art.

[0134] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and implementation modes. They can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the illustrations shown and described herein.

[0135] The apparatus, computer device, non-volatile computer storage medium and method provided in the embodiments of this specification correspond to each other, and therefore, the apparatus, computer device and non-volatile computer storage medium also have similar beneficial technical effects as the corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the corresponding apparatus, computer device and non-volatile computer storage medium will not be repeated here.

[0136] Those skilled in the art also know that, in addition to implementing the controller in a purely computer-readable program code, the controller can be made to implement the same function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered as a hardware component, and the devices for implementing various functions included therein can also be considered as structures within the hardware component. Or even, the devices for implementing various functions can be considered as both software units for implementing the method and structures within the hardware component.

[0137] The systems, devices or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. For the convenience of description, the above devices are described separately by functions in various units. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in the same or more software and / or hardware.

[0138] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification may be in the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the embodiments of this specification may be in the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0139] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0140] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0141] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0142] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0143] The specification may be described in the general context of computer-executable instructions executed by a computer, such as program units. Generally, program units include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communications network. In a distributed computing environment, program units may be located in local and remote computer storage media, including storage devices.

[0144] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0145] The above description is only an embodiment of this specification and is not intended to limit one or more embodiments of this specification. For those skilled in the art, one or more embodiments of this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included in the scope of the claims of one or more embodiments of this specification.

Claims

1. A brain magnetic resonance image super-resolution reconstruction method based on an implicit representation model, characterized in that: The following steps are involved: Acquire anisotropic brain magnetic resonance images of different slice thicknesses and divide them into training data and test data; Preprocessing the anisotropic brain magnetic resonance image; Processing the preprocessed training data to generate training samples; Using the training samples to train a super-resolution reconstruction model based on a latent representation model; Input the preprocessed test data into a trained model, output a reconstructed image slice image, and stack them to obtain a three-dimensional image; The obtained three-dimensional images are then averaged to finally obtain an isotropic brain magnetic resonance image.

2. The method for super-resolution reconstruction of brain magnetic resonance images based on an implicit representation model as claimed in claim 1, characterized in that: The step of preprocessing the anisotropic brain magnetic resonance image comprises: performing bias field correction on the anisotropic brain magnetic resonance image; The pixel values ​​of the anisotropic brain magnetic resonance image are subjected to maximum and minimum normalization processing.

3. The method for super-resolution reconstruction of brain magnetic resonance images based on an implicit representation model as claimed in claim 1, characterized in that: The step of processing the preprocessed training data to generate training samples includes: Extracting high-resolution image blocks according to preset sizes from high-resolution intra-layer slices; According to the ratio of the inter-layer and intra-layer voxel resolutions, Gaussian blurring and one-dimensional downsampling are performed on the high-resolution image block, and the downsampling scale is the ratio; The downsampled image blocks are upsampled to obtain low-resolution image blocks of the same size as the high-resolution image blocks, and constitute a set of training samples.

4. The method for super-resolution reconstruction of brain magnetic resonance images based on an implicit representation model as claimed in claim 3, characterized in that: The super-resolution reconstruction model based on the hidden representation model includes an image feature extraction module, a learnable coordinate encoding module and a decoding module; wherein, The image feature extraction module is used to extract image features from image blocks; The learnable coordinate encoding module is used to extract coordinate encoding from image block coordinates; The decoding module is used to decode the fusion feature of the image feature and the coordinate code.

5. The method for super-resolution reconstruction of brain magnetic resonance images based on an implicit representation model as claimed in claim 4, characterized in that: The image feature extraction module adopts an EDSR-based network structure, which includes a convolutional layer, a residual block and a jump connection. The residual block sequentially includes a convolution, a linear rectifier unit, a convolution and a constant scaling layer. The jump connection connects the feature map output by the first convolutional layer with the output feature map of the penultimate convolutional layer, and sends the connected features to the last convolutional layer.

6. The method for super-resolution reconstruction of brain magnetic resonance images based on an implicit representation model as claimed in claim 4, characterized in that: The learnable coordinate encoding module includes multiple fully connected layers, whose input feature dimensions correspond to the coordinates of each pixel in the image. The first fully connected layer maps the input features to a first preset dimensional space and performs a nonlinear transformation through an activation function. The middle fully connected layer further expands the multi-dimensional features of the previous fully connected layer to more dimensions and is also followed by an activation function. The penultimate fully connected layer maintains the multi-dimensional features of the previous fully connected layer and is also followed by an activation function. The last connected layer maps the multi-dimensional features to the final output feature dimensions.

7. The method for super-resolution reconstruction of brain magnetic resonance images based on an implicit representation model as claimed in claim 4, characterized in that: The decoding module includes multiple fully connected layers, each of which is followed by an activation function. A residual connection is also introduced in the decoding module to add the input information to the output of the intermediate activation layer.

8. The method for super-resolution reconstruction of brain magnetic resonance images based on an implicit representation model as claimed in claim 4, characterized in that: The loss function for training the super-resolution reconstruction model based on the implicit representation model using the training samples adopts a hybrid loss function that integrates content loss and brain tissue edge perception loss, and the formula is defined as follows: Among them, y represents the real image, Represents the prediction result of super-resolution reconstruction, L c Indicates content loss, L eg represents the edge perception loss of brain tissue, and weights λ1 and λ2 are used to balance different loss function terms; L c , L eg The definition is as follows: Among them, C h and C w Represents the horizontal and vertical edge detection Prewitt operators, represents the convolution operation, M y and Represent the real graph of y after the Prewitt operator and Gradient magnitude map of the prediction image.

9. A brain magnetic resonance image super-resolution reconstruction system based on an implicit representation model, implementing the method according to any one of claims 1 to 8, characterized in that: It includes image acquisition module, image preprocessing module, training sample generation module, model training module, model testing module and image processing module; among them, The image acquisition module is used to acquire anisotropic brain magnetic resonance images of different layer thicknesses, which are divided into training data and test data; The image preprocessing module is used to preprocess the anisotropic brain magnetic resonance image; The training sample generating module is used to process the pre-processed training data to generate training samples; The model training module is used to train a super-resolution reconstruction model based on a latent representation model using the training samples; The model testing module is used to input the preprocessed test data into the trained model, output the reconstructed image slice images, and stack them to obtain a three-dimensional image; The image processing module is used to average the obtained three-dimensional images to finally obtain an isotropic brain magnetic resonance image.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images