A method and device for reconstructing high-resolution MRI sequences

Abstract

The present invention discloses a reconstruction method and device of a high-level inter-resolution MRI sequence, and relates to the field of magnetic resonance imaging technology. The method comprises: collecting image data pairs composed of low-level and high-level inter-resolution MRI sequences with consistent real spatial positions of image matrices, and establishing a training data set; constructing a reconstruction neural network, and constructing a deep learning network loss of the reconstruction neural network; inputting the training data set into the reconstruction neural network for deep learning training, iteratively updating the weights of the reconstruction neural network until the deep learning network loss tends to be stable, so as to obtain an inter-level resolution reconstruction model; obtaining a low-level inter-resolution MRI sequence to be reconstructed, and inputting the low-level inter-resolution MRI sequence to be reconstructed into the inter-level resolution reconstruction model to obtain a high-level inter-resolution MRI sequence. The present invention solves the problem of low efficiency in obtaining high-level inter-resolution MRI sequences in the prior art.

Description

Technical Field

[0001] The present invention relates to the technical field of magnetic resonance imaging, and in particular to a method and device for reconstructing a high-resolution MRI sequence. Background Art

[0002] Magnetic resonance imaging (MRI) is a modern medical imaging technique that is widely used in hospitals and clinics as an important diagnostic tool. It uses magnetic fields and harmless radio waves to create precise images of internal human tissues and is particularly good at diagnosing abnormalities or diseases of the brain and spine.

[0003] MRI sequences (images) are very useful for medical experts because they provide detailed information about tissue structure, helping doctors determine the type and location of lesions. However, because lesions in some organs are very small and complex, MRI sequences with low inter-slice resolution may not provide enough details, which has given rise to the need for high inter-slice resolution MRI sequences.

[0004] Currently, there are two main techniques for obtaining MRI sequences with high inter-layer resolution: 3D MRI scanning and multi-layer 2D MRI scanning. In 3D MRI scanning, the entire imaging area is first excited, and then spatial encoding is used to obtain a three-dimensional image. Multi-layer 2D MRI scanning acquires images by scanning layer by layer, first selecting a specific layer of interest, and then positioning it through gradients in the other two directions.

[0005] However, these scans can take a long time and require the patient to remain still for a long time, which can cause discomfort and be inefficient. Summary of the invention

[0006] In view of this, an object of the present invention is to provide a method and device for reconstructing a high-resolution MRI sequence, aiming to solve the problem of low efficiency in obtaining a high-resolution MRI sequence in the prior art.

[0007] The embodiment of the present invention is implemented as follows:

[0008] A method for reconstructing a high-resolution MRI sequence, the method comprising:

[0009] Acquire image data pairs consisting of a low-layer inter-resolution MRI sequence and a high-layer inter-resolution MRI sequence with consistent real spatial positions of the image matrix, and establish a training data set based on multiple groups of the image data pairs;

[0010] Constructing an inter-layer resolution reconstruction neural network composed of a plurality of iterative units, and constructing a deep learning network loss of the reconstruction neural network;

[0011] Inputting the training data set into the reconstruction neural network for deep learning training, iteratively updating the weights of the reconstruction neural network until the loss of the deep learning network tends to be stable, so as to obtain an inter-layer resolution reconstruction model;

[0012] Acquire a low-inter-layer resolution MRI sequence to be reconstructed, and input the low-inter-layer resolution MRI sequence to be reconstructed into the inter-layer resolution reconstruction model to obtain a high-inter-layer resolution MRI sequence;

[0013] The step of acquiring an image data pair consisting of a low-inter-layer resolution MRI sequence and a high-inter-layer resolution MRI sequence with consistent real spatial positions of the image matrix also includes:

[0014] Acquire the high-resolution MRI sequence obtained by performing a multi-layer 2D MRI scan or a 3D MRI scan on the same target person;

[0015] Performing a degradation operation on the high-level inter-layer resolution MRI sequence to obtain a matching low-level inter-layer resolution MRI sequence;

[0016] The step of performing a degradation operation on the high-level inter-layer resolution MRI sequence to obtain the matched low-level inter-layer resolution MRI sequence comprises:

[0017] Decomposing the high-resolution MRI sequence at multiple scales, and degrading the high-resolution MRI sequence at each scale using different degradation methods;

[0018] Reconstructing the high-level inter-resolution MRI sequence of each scale after degradation to obtain the low-level inter-resolution MRI sequence;

[0019] The degradation methods include noise addition, blur processing, layered compression and detail loss, texture disturbance, and random occlusion and deletion.

[0020] Furthermore, in the above-mentioned method for reconstructing a high-resolution MRI sequence between layers, the step of acquiring an image data pair consisting of a low-resolution MRI sequence between layers and a high-resolution MRI sequence between layers having the same real spatial position of the image matrix further includes:

[0021] Obtain 2D MRI scans of the same target person in three different planes to obtain high-level internal resolution MRI images in different planes;

[0022] A degradation operation is performed on the high-level intra-resolution MRI image to obtain the low-level inter-level resolution MRI sequence corresponding to the plane.

[0023] Furthermore, in the above-mentioned method for reconstructing a high-level inter-resolution MRI sequence, the step of performing a degradation operation on the high-level intra-resolution MRI image to obtain the low-level inter-resolution MRI sequence corresponding to the plane comprises:

[0024] The high-level intra-resolution MRI image is normalized, and the two-dimensional MRI data pixels of the high-level intra-resolution MRI image are blurred, the image resolution is reduced, noise is added, and ringing effect is added in sequence to downsample the high-level intra-resolution MRI image to obtain a low-level intra-resolution MRI image, thereby obtaining the low-inter-layer resolution MRI sequence.

[0025] Furthermore, in the above-mentioned high-resolution MRI sequence reconstruction method, the step of adding noise includes:

[0026] generating a 3D noise field, and superimposing the 3D noise field onto the volume of the high inter-resolution MRI sequence;

[0027] The fuzzy processing steps include:

[0028] Convolving the volume of the high inter-resolution MRI sequence using a 3D Gaussian filter;

[0029] The steps of layered compression and detail loss include:

[0030] The high inter-resolution MRI sequence is divided into layers of different depths along the depth direction, and different degrees of compression and detail loss are applied to each layer.

[0031] Furthermore, in the above-mentioned reconstruction method of the high-resolution MRI sequence, the texture perturbation step comprises:

[0032] Calculating a normal of a volume of a high inter-resolution MRI sequence and adding a random perturbation in the direction of the normal;

[0033] The random occlusion and missing steps include:

[0034] A number of occlusion regions are randomly generated, and the occlusion regions are applied to the volume of the high-resolution MRI sequence.

[0035] Furthermore, in the above-mentioned reconstruction method of the high-resolution MRI sequence, the iteration unit includes a data consistency unit and a regularization unit based on a convolutional neural network.

[0036] Furthermore, in the above-mentioned high-resolution MRI sequence reconstruction method, the deep learning network loss includes one or more of pixel feature loss, image feature loss, adversarial generation loss, and k-space feature loss.

[0037] Another object of the present invention is to provide a device for reconstructing a high-resolution MRI sequence, the device comprising:

[0038] An acquisition module, used for acquiring image data pairs consisting of a low-layer inter-resolution MRI sequence and a high-layer inter-resolution MRI sequence having consistent real spatial positions of an image matrix, and establishing a training data set based on a plurality of sets of the image data pairs;

[0039] A construction module, used to construct an inter-layer resolution reconstruction neural network composed of multiple iteration units, and to construct a deep learning network loss of the reconstruction neural network;

[0040] A training module, used for inputting the training data set into the reconstruction neural network for deep learning training, and iteratively updating the weights of the reconstruction neural network until the loss of the deep learning network tends to be stable, so as to obtain an inter-layer resolution reconstruction model;

[0041] The reconstruction module is used to obtain a low-inter-layer resolution MRI sequence to be reconstructed, and input the low-inter-layer resolution MRI sequence to be reconstructed into the inter-layer resolution reconstruction model to obtain a high-inter-layer resolution MRI sequence.

[0042] Another object of the present invention is to provide a readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the steps of any one of the methods described above.

[0043] Another object of the present invention is to provide an electronic 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 program.

[0044] The present invention collects image data pairs consisting of low-layer resolution MRI sequences and high-layer resolution MRI sequences with consistent real spatial positions of image matrices, and establishes training data sets according to multiple sets of image data pairs; constructs an interlayer resolution reconstruction neural network consisting of multiple iteration units, and constructs a deep learning network loss of the reconstruction neural network; inputs the training data set into the reconstruction neural network for deep learning training, iteratively updates the weights of the reconstruction neural network until the deep learning network loss tends to be stable, so as to obtain an interlayer resolution reconstruction model; obtains a low-layer resolution MRI sequence to be reconstructed, and inputs the low-layer resolution MRI sequence to be reconstructed into the interlayer resolution reconstruction model to obtain a high-layer resolution MRI sequence. Since the interlayer resolution reconstruction model has mastered the generation logic of the high-layer resolution MRI sequence, the high-layer resolution MRI sequence can be directly obtained through the low-layer resolution MRI sequence, and the efficiency is greatly improved. The problem of low efficiency in obtaining high-layer resolution MRI sequences in the prior art is solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A schematic flow chart of a method for reconstructing a high-resolution MRI sequence provided by an embodiment of the present invention;

[0046] Figure 2 It is a schematic diagram of the structure of a reconstruction neural network in a method for reconstructing a high-resolution MRI sequence in one embodiment of the present invention;

[0047] Figure 3 It is a structural block diagram of a reconstruction device for a high-inter-resolution MRI sequence in a third embodiment of the present invention.

[0048] The following specific implementation manner will further illustrate the present invention in conjunction with the above-mentioned drawings. DETAILED DESCRIPTION

[0049] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Several embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0050] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0051] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0052] The following will describe in detail how to solve the problem of low efficiency in acquiring high-resolution MRI sequences in the prior art in combination with specific embodiments and drawings.

[0053] Embodiment 1

[0054] See also Figure 1 , shown is a reconstruction method of a high-level inter-resolution MRI sequence in a first embodiment of the present invention, the method comprising steps S10 to S13.

[0055] Step S10, collecting image data pairs consisting of a low-layer resolution MRI sequence and a high-layer resolution MRI sequence with consistent real spatial positions of the image matrix, and establishing a training data set based on multiple groups of the image data pairs.

[0056] Among them, image data pairs of mutually matching MRI sequences of low-layer resolution and high-layer resolution are collected, and a plurality of sets of image data pairs are used to establish a training data set, and the training data set is used for learning and training of the neural network. Specifically, mutual matching means that the real spatial position of the image matrix is ​​consistent. In the specific implementation, the corresponding matching low-layer resolution MRI sequences and high-layer resolution MRI sequences can be obtained respectively, and an image database can be established, so that the corresponding MRI sequence can be found and obtained from the data when needed.

[0057] It should also be noted that, in practice, in magnetic resonance imaging (MRI), in-plane resolution and slice thickness or inter-slice resolution are two important concepts, which describe the resolution of different aspects of the image. For example, in a 2D MRI scan of the transverse (axial) plane, when the sagittal and coronal planes are viewed through software, the image resolution of these two planes is called inter-slice resolution, and the image resolution of the axial plane is called in-plane resolution. Specifically, in-plane resolution refers to the resolution within each MRI image slice, that is, the resolution of the image on the cross-section (usually the cross section or coronal plane). It is determined by the pixel size, which is expressed as the size of each pixel in real space. The in-plane resolution is determined by the FOV (Field of View) and the matrix size. Thinner slices provide higher inter-slice resolution. In the embodiment of the present invention, it is mainly for high-resolution MRI sequences to be reconstructed.

[0058] Step S11, constructing an inter-layer resolution reconstruction neural network composed of multiple iterative units, and constructing a deep learning network loss of the reconstruction neural network.

[0059] Among them, a neural network for learning and training is constructed, and the backbone network can input five-dimensional image data, such as Figure 2As shown in the figure, the backbone network is mainly composed of multiple iteration units. Specifically, multiple iteration units form an iteration unit group. Each iteration unit in the group has a data consistency (DC) unit and a regularization unit based on a convolutional neural network. In the figure, FFT is a forward Fourier transform and IFFT is an inverse Fourier transform. The iteration unit group generates a reconstructed image with a high-level inter-resolution from the k-space data with a low-level inter-resolution. The multi-channel input of each iteration unit includes not only the output of the previous iteration unit, but also the output of the optional N previous iteration units. The multi-channel input of the regularization unit is an optional M-image, which is composed of multiple convolutional layers and LReLu activation layers.

[0060] In a specific implementation, the deep learning network loss is composed of one or more pixel feature losses, image feature losses, adversarial generation losses, and k-space feature losses, wherein pixel feature losses include L1loss, L2loss, and SmoothL1loss; image feature losses include SSIM loss, PSNR loss, and losses using public pre-trained network feature maps; pre-trained networks include but are not limited to neural networks trained based on CNN, Transformer, and Diffusion; adversarial generation losses are composed of trainable discriminator networks and GAN losses; the discriminant network can be a VGG network, or a basic FCN network, or a U-net network based on spectral normalization and batch normalization, or a VIT segmentation network based on Transformer; and the GAN loss can be a basic GAN loss, a WGAN loss, or an RGAN loss.

[0061] Step S12, inputting the training data set into the reconstruction neural network for deep learning training, iteratively updating the weights of the reconstruction neural network until the loss of the deep learning network tends to be stable, so as to obtain an inter-layer resolution reconstruction model.

[0062] Among them, the matrix of low-layer resolution MRI sequences in the training set is used as the input of the network, and the results of high-layer resolution are output. The loss is calculated using the data of high-layer resolution paired with the low-layer resolution MRI sequences, and the network weights are iteratively updated through stochastic gradient descent or Adam optimizer until the loss of the deep learning network stabilizes.

[0063] Step S13, obtaining a low-inter-layer resolution MRI sequence to be reconstructed, and inputting the low-inter-layer resolution MRI sequence to be reconstructed into the inter-layer resolution reconstruction model to obtain a high-inter-layer resolution MRI sequence.

[0064] Among them, the inter-layer resolution reconstruction model masters the generation logic of high-layer inter-resolution MRI sequences. When a low-layer inter-resolution MRI sequence is obtained, it can be reconstructed to obtain the corresponding high-layer inter-resolution MRI sequence. The low-layer inter-resolution MRI sequence does not require 3D MRI scanning and multi-layer 2D MRI scanning, but only requires a simple 2D MRI scan, thereby avoiding the long scanning process.

[0065] In summary, the reconstruction method of the high-level inter-resolution MRI sequence in the above-mentioned embodiment of the present invention collects image data pairs consisting of low-level inter-resolution MRI sequences and high-level inter-resolution MRI sequences with consistent real spatial positions of the image matrix, and establishes a training data set based on multiple sets of image data pairs; constructs an inter-level resolution reconstruction neural network consisting of multiple iteration units, and constructs a deep learning network loss of the reconstruction neural network; inputs the training data set into the reconstruction neural network for deep learning training, iteratively updates the weights of the reconstruction neural network until the deep learning network loss tends to be stable, so as to obtain an inter-level resolution reconstruction model; obtains a low-level inter-resolution MRI sequence to be reconstructed, and inputs the low-level inter-resolution MRI sequence to be reconstructed into the inter-level resolution reconstruction model to obtain a high-level inter-resolution MRI sequence. Since the inter-level resolution reconstruction model has mastered the generation logic of the high-level inter-resolution MRI sequence, the high-level inter-resolution MRI sequence can be directly obtained through the low-level inter-resolution MRI sequence, and the efficiency is greatly improved. The problem of low efficiency in obtaining high-level inter-resolution MRI sequences in the prior art is solved.

[0066] Embodiment 2

[0067] This embodiment also proposes a method for reconstructing a high-resolution MRI sequence. The method for reconstructing a high-resolution MRI sequence in this embodiment is different from the method for reconstructing a high-resolution MRI sequence in the first embodiment in that:

[0068] Before step S10, the method further includes:

[0069] Obtain 2D MRI scans of the same target person in three different planes to obtain high-level internal resolution MRI images in different planes;

[0070] A degradation operation is performed on the high-level intra-resolution MRI image to obtain the low-level inter-level resolution MRI sequence corresponding to the plane.

[0071] Among them, the high-level intra-resolution MRI image is normalized, and the two-dimensional MRI data pixels of the high-level intra-resolution MRI image are blurred, the image resolution is reduced, noise is added, and ringing effect is added in sequence to downsample the high-level intra-resolution MRI image to obtain a low-level intra-resolution MRI image, thereby obtaining the low-inter-layer resolution MRI sequence.

[0072] Specifically, the collected high-level internal resolution MRI images are normalized by analyzing the statistical histogram of the pixel values ​​of the image. Optionally, the pixel blurring of the two-dimensional MRI data includes Gaussian blur (isotropic, anisotropic), motion blur, disk blur, image shadow, tilt-shift blur, path blur, scene blur, rotation blur, and real blur simulation based on deep learning.

[0073] Optionally, methods for reducing pixel resolution of two-dimensional MRI data include monolinear interpolation, bilinear interpolation, bicubic interpolation, etc.;

[0074] Optionally, the two-dimensional MRI data pixel adding noise processing includes Gaussian noise, Rice noise, salt and pepper noise, Poisson noise, and a noise generation network obtained from a real two-dimensional data image based on deep learning technology;

[0075] The downsampling method may optionally perform the above steps once or multiple times to obtain a low intra-layer resolution MRI image.

[0076] Among them, the collected high-level intra-resolution MRI images are sharpened to obtain MRI images with higher intra-level resolution, and then a matching data pair is obtained. Specifically, the sharpening process includes a differentiation method, a high-pass filtering method, and the differentiation method includes an addible gradient method, a Sobel operator method, and a Laplace operator method. The sharpening method can optionally perform the above steps once or multiple times. After obtaining the data pairs of low-level intra-resolution and high-level intra-resolution images, the data pairs of high-level inter-resolution MRI sequences and low-level inter-resolution MRI sequences can be obtained.

[0077] In addition, in some optional embodiments of the present invention, before step S10, the following steps are further included:

[0078] Acquire the high-resolution MRI sequence obtained by performing a multi-layer 2D MRI scan or a 3D MRI scan on the same target person;

[0079] A degradation operation is performed on the high-inter-layer resolution MRI sequence to obtain a matched low-inter-layer resolution MRI sequence.

[0080] The step of performing a degradation operation on the high-level inter-resolution MRI image to obtain a matching low-level inter-resolution MRI image comprises:

[0081] Decomposing the high-resolution MRI sequence at multiple scales, and degrading the high-resolution MRI sequence at each scale using different degradation methods;

[0082] The degraded high-inter-layer resolution MRI sequence of each scale is reconstructed to obtain the low-inter-layer resolution MRI sequence.

[0083] Specifically, the degradation methods include noise addition, blur processing, layered compression and detail loss, texture disturbance, and random occlusion and deletion;

[0084] Optionally, the noise adding step includes:

[0085] generating a 3D noise field, and superimposing the 3D noise field onto the volume of the high inter-resolution MRI sequence;

[0086] Specifically, a 3D noise field, such as Perlin noise or simplex noise, is generated, and the 3D noise field is superimposed on the volume of the 3D image of the high-resolution MRI sequence, so that the noise has spatial gradient and correlation;

[0087] Optionally, the obfuscation steps include:

[0088] Convolving the volume of the high inter-resolution MRI sequence using a 3D Gaussian filter;

[0089] Specifically, a 3D Gaussian filter is used to perform a convolution operation on the 3D image volume of the high-resolution MRI sequence, and the 3D Gaussian filter is applied to ensure that the blurring effect is continuous in space, thereby obtaining a more realistic degradation result compared to 2D blurring, wherein the standard deviation of the filter is adjusted to control the blurring degree;

[0090] Optional, layered compression with detail loss steps include:

[0091] Dividing the high-resolution MRI sequence into layers of different depths along the depth direction, and applying different degrees of compression and detail loss to each layer;

[0092] Specifically, the 3D images of the high-resolution MRI sequence are divided into layers of different depths along the depth direction, and different degrees of compression and detail loss are applied to each layer. The deeper layers are compressed at a higher rate and lose more details, while the shallower layers retain more details.

[0093] Optionally, the texture perturbation step includes:

[0094] Calculating a normal of a volume of a high inter-resolution MRI sequence and adding a random perturbation in the direction of the normal;

[0095] Specifically, the normal of the 3D image volume of the high-resolution MRI sequence is calculated, and random perturbations are added in the normal direction to introduce texture perturbations so that the texture no longer looks smooth or consistent, wherein the intensity of the perturbations is adjusted to control the degree of texture deformation;

[0096] Optional, random occlusion and deletion steps include:

[0097] randomly generating a number of occlusion regions, and applying the occlusion regions to the volume of the high inter-resolution MRI sequence;

[0098] Specifically, several occlusion regions are randomly generated, and the occlusion regions are applied to the 3D image volume of the high-resolution MRI sequence. Occlusion objects of random shapes are introduced into the 3D image or some voxels are removed to simulate data loss. In images with thicker layers, some lesions are easily made unclear. This method can obtain a more realistic degraded image. It is difficult for a 2D degradation scheme to achieve this degradation effect. Several occlusion regions or shapes, such as spheres and cubes, are randomly generated. These occlusion regions are applied to the 3D image volume so that the data of the corresponding parts is 0 or other specified values. The number and size of the occlusion regions are controlled to adjust the degree of occlusion.

[0099] In addition, the acquired high-resolution MRI sequence is sharpened to obtain a higher-resolution MRI sequence, which is paired with the previous low-resolution MRI sequence. The sharpening process includes differential method and high-pass filter method. The differential method includes an additive gradient method, Sobel operator method, and Laplace operator method. The sharpening process can be performed once or multiple times.

[0100] In summary, the reconstruction method of the high-level inter-resolution MRI sequence in the above-mentioned embodiment of the present invention collects image data pairs consisting of low-level inter-resolution MRI sequences and high-level inter-resolution MRI sequences with consistent real spatial positions of the image matrix, and establishes a training data set based on multiple sets of image data pairs; constructs an inter-level resolution reconstruction neural network consisting of multiple iteration units, and constructs a deep learning network loss of the reconstruction neural network; inputs the training data set into the reconstruction neural network for deep learning training, iteratively updates the weights of the reconstruction neural network until the deep learning network loss tends to be stable, so as to obtain an inter-level resolution reconstruction model; obtains a low-level inter-resolution MRI sequence to be reconstructed, and inputs the low-level inter-resolution MRI sequence to be reconstructed into the inter-level resolution reconstruction model to obtain a high-level inter-resolution MRI sequence. Since the inter-level resolution reconstruction model has mastered the generation logic of the high-level inter-resolution MRI sequence, the high-level inter-resolution MRI sequence can be directly obtained through the low-level inter-resolution MRI sequence, and the efficiency is greatly improved. The problem of low efficiency in obtaining high-level inter-resolution MRI sequences in the prior art is solved.

[0101] Embodiment 3

[0102] See also Figure 3 , shown is a reconstruction device for a high-resolution MRI sequence proposed in a third embodiment of the present invention, the device comprising:

[0103] An acquisition module 100 is used to acquire image data pairs consisting of a low-layer resolution MRI sequence and a high-layer resolution MRI sequence with consistent real spatial positions of an image matrix, and to establish a training data set based on a plurality of sets of the image data pairs;

[0104] A construction module 200 is used to construct an inter-layer resolution reconstruction neural network composed of multiple iteration units, and to construct a deep learning network loss of the reconstruction neural network;

[0105] A training module 300 is used to input the training data set into the reconstruction neural network for deep learning training, and iteratively update the weights of the reconstruction neural network until the loss of the deep learning network tends to be stable, so as to obtain an inter-layer resolution reconstruction model;

[0106] The reconstruction module 400 is used to obtain a low-inter-layer resolution MRI sequence to be reconstructed, and input the low-inter-layer resolution MRI sequence to be reconstructed into the inter-layer resolution reconstruction model to obtain a high-inter-layer resolution MRI sequence.

[0107] Furthermore, the above-mentioned high-resolution MRI sequence reconstruction device further includes:

[0108] The first scanning module is used to obtain 2D MRI scans of the same target person in three different planes to obtain high-level internal resolution MRI images in different planes;

[0109] A degradation operation is performed on the high-level intra-resolution MRI image to obtain the low-level inter-level resolution MRI sequence corresponding to the plane.

[0110] Furthermore, the above-mentioned high-resolution MRI sequence reconstruction device further comprises:

[0111] A second scanning module is used to obtain the high-resolution MRI sequence between layers obtained by performing a multi-layer 2D MRI scan or a 3D MRI scan on the same target person;

[0112] A degradation operation is performed on the high-inter-layer resolution MRI sequence to obtain a matched low-inter-layer resolution MRI sequence.

[0113] Furthermore, in the above-mentioned high-resolution MRI sequence reconstruction device, the first scanning module is specifically used for:

[0114] The high-level intra-resolution MRI image is normalized, and the two-dimensional MRI data pixels of the high-level intra-resolution MRI image are blurred, the image resolution is reduced, noise is added, and ringing effect is added in sequence to downsample the high-level intra-resolution MRI image to obtain a low-level intra-resolution MRI image, thereby obtaining the low-inter-layer resolution MRI sequence.

[0115] Furthermore, in the above-mentioned high-resolution MRI sequence reconstruction device, the second scanning module is specifically used for:

[0116] Decomposing the high-resolution MRI sequence at multiple scales, and degrading the high-resolution MRI sequence at each scale using different degradation methods;

[0117] Reconstructing the high-level inter-resolution MRI sequence of each scale after degradation to obtain the low-level inter-resolution MRI sequence;

[0118] The degradation methods include noise addition, blur processing, layered compression and detail loss, texture disturbance, and random occlusion and deletion.

[0119] Furthermore, in the above-mentioned high-resolution MRI sequence reconstruction device, the step of adding noise includes:

[0120] generating a 3D noise field, and superimposing the 3D noise field onto the volume of the high inter-resolution MRI sequence;

[0121] The fuzzy processing steps include:

[0122] Convolving the volume of the high inter-resolution MRI sequence using a 3D Gaussian filter;

[0123] The steps of layered compression and detail loss include:

[0124] Dividing the high-resolution MRI sequence into layers of different depths along the depth direction, and applying different degrees of compression and detail loss to each layer;

[0125] The texture perturbation step comprises:

[0126] Calculating a normal of a volume of a high inter-resolution MRI sequence and adding a random perturbation in the direction of the normal;

[0127] The random occlusion and missing steps include:

[0128] A number of occlusion regions are randomly generated, and the occlusion regions are applied to the volume of the high-resolution MRI sequence.

[0129] Furthermore, in the above-mentioned reconstruction device for high-resolution MRI sequence, the iteration unit includes a data consistency unit and a regularization unit based on a convolutional neural network.

[0130] Furthermore, in the above-mentioned high-resolution MRI sequence reconstruction device, the deep learning network loss includes one or more of pixel feature loss, image feature loss, adversarial generation loss, and k-space feature loss.

[0131] The functions or operation steps implemented when the above modules are executed are substantially the same as those in the above method embodiments, and will not be repeated here.

[0132] Embodiment 4

[0133] Another aspect of the present invention further provides a readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the steps of the method described in any one of the above-mentioned embodiments 1 to 2 are implemented.

[0134] Embodiment 5

[0135] Another aspect of the present invention provides an electronic device, which includes 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 method described in any one of the above-mentioned embodiments 1 to 2 when executing the program.

[0136] The technical features of the above embodiments can 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.

[0137] Those skilled in the art will appreciate that the logic and / or steps represented in the flowchart or otherwise described herein, for example, may be considered as an ordered list of executable instructions for implementing logical functions, and may be specifically implemented in any computer-readable storage medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For purposes of this specification, a "computer-readable storage medium" may be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0138] More specific examples (a non-exhaustive list) of computer-readable storage media include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable storage medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0139] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or a combination thereof: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0140] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0141] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for reconstructing a high-resolution MRI sequence, characterized in that: The method comprises: Acquire image data pairs consisting of a low-layer inter-resolution MRI sequence and a high-layer inter-resolution MRI sequence with consistent real spatial positions of the image matrix, and establish a training data set based on multiple groups of the image data pairs; Constructing an inter-layer resolution reconstruction neural network composed of a plurality of iterative units, and constructing a deep learning network loss of the reconstruction neural network; Inputting the training data set into the reconstruction neural network for deep learning training, iteratively updating the weights of the reconstruction neural network until the loss of the deep learning network tends to be stable, so as to obtain an inter-layer resolution reconstruction model; Acquire a low-inter-layer resolution MRI sequence to be reconstructed, and input the low-inter-layer resolution MRI sequence to be reconstructed into the inter-layer resolution reconstruction model to obtain a high-inter-layer resolution MRI sequence; The step of acquiring an image data pair consisting of a low-inter-layer resolution MRI sequence and a high-inter-layer resolution MRI sequence with consistent real spatial positions of the image matrix also includes: Acquire the high-resolution MRI sequence obtained by performing a multi-layer 2D MRI scan or a 3D MRI scan on the same target person; Performing a degradation operation on the high-level inter-layer resolution MRI sequence to obtain a matching low-level inter-layer resolution MRI sequence; The step of performing a degradation operation on the high-level inter-layer resolution MRI sequence to obtain the matched low-level inter-layer resolution MRI sequence comprises: Decomposing the high-resolution MRI sequence at multiple scales, and degrading the high-resolution MRI sequence at each scale using different degradation methods; Reconstructing the high-level inter-resolution MRI sequence of each scale after degradation to obtain the low-level inter-resolution MRI sequence; The degradation methods include noise addition, blurring, layered compression and detail loss, texture disturbance, and random occlusion and deletion; The step of adding noise includes: generating a 3D noise field and superimposing the 3D noise field onto the volume of the high inter-resolution MRI sequence so that the noise is spatially gradual and correlated; The fuzzy processing steps include: convolving the volume of the high inter-resolution MRI sequence using a 3D Gaussian filter so that the blurring effect is spatially continuous; The steps of layered compression and detail loss include: Dividing the high-resolution MRI sequence into layers of different depths along the depth direction, and applying different degrees of compression and detail loss to each layer; The texture perturbation step comprises: Calculating a normal of a volume of a high inter-resolution MRI sequence and adding a random perturbation in the direction of the normal; The random occlusion and missing steps include: A number of occlusion regions are randomly generated, and the occlusion regions are applied to the volume of the high-resolution MRI sequence.

2. The method for reconstructing a high-resolution MRI sequence according to claim 1, characterized in that: The step of acquiring an image data pair consisting of a low-inter-layer resolution MRI sequence and a high-inter-layer resolution MRI sequence with consistent real spatial positions of the image matrix also includes: Obtain 2D MRI scans of the same target person in three different planes to obtain high-level internal resolution MRI images in different planes; A degradation operation is performed on the high-level intra-resolution MRI image to obtain the low-level inter-level resolution MRI sequence corresponding to the plane.

3. The method for reconstructing a high-resolution MRI sequence according to claim 2, characterized in that: The step of performing a degradation operation on the high-level intra-resolution MRI image to obtain the low-level inter-level resolution MRI sequence corresponding to the plane comprises: The high-level intra-resolution MRI image is normalized, and the two-dimensional MRI data pixels of the high-level intra-resolution MRI image are blurred, the image resolution is reduced, noise is added, and ringing effect is added in sequence to downsample the high-level intra-resolution MRI image to obtain a low-level intra-resolution MRI image, thereby obtaining the low-inter-layer resolution MRI sequence.

4. The method for reconstructing a high-resolution MRI sequence according to claim 1, characterized in that: The iteration unit includes a data consistency unit and a regularization unit based on a convolutional neural network.

5. The method for reconstructing a high-resolution MRI sequence according to claim 1, characterized in that: The deep learning network loss includes one or more of pixel feature loss, image feature loss, adversarial generation loss, and k-space feature loss.

6. A reconstruction device for high-resolution MRI sequences, characterized in that: A method for reconstructing a high-resolution MRI sequence according to any one of claims 1 to 5, the device comprising: An acquisition module, used for acquiring image data pairs consisting of a low-layer inter-resolution MRI sequence and a high-layer inter-resolution MRI sequence having consistent real spatial positions of an image matrix, and establishing a training data set based on a plurality of sets of the image data pairs; A construction module, used to construct an inter-layer resolution reconstruction neural network composed of multiple iteration units, and to construct a deep learning network loss of the reconstruction neural network; A training module, used for inputting the training data set into the reconstruction neural network for deep learning training, and iteratively updating the weights of the reconstruction neural network until the loss of the deep learning network tends to be stable, so as to obtain an inter-layer resolution reconstruction model; The reconstruction module is used to obtain a low-inter-layer resolution MRI sequence to be reconstructed, and input the low-inter-layer resolution MRI sequence to be reconstructed into the inter-layer resolution reconstruction model to obtain a high-inter-layer resolution MRI sequence.

7. A readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

8. An electronic device, characterized in that: The method comprises 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 method according to any one of claims 1 to 5 when executing the program.

Images