CN117576240A - Magnetic resonance image reconstruction method based on double-domain transducer - Google Patents
Magnetic resonance image reconstruction method based on double-domain transducer Download PDFInfo
- Publication number
- CN117576240A CN117576240A CN202311586159.1A CN202311586159A CN117576240A CN 117576240 A CN117576240 A CN 117576240A CN 202311586159 A CN202311586159 A CN 202311586159A CN 117576240 A CN117576240 A CN 117576240A
- Authority
- CN
- China
- Prior art keywords
- image
- space data
- magnetic resonance
- reconstructed
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000012545 processing Methods 0.000 claims abstract description 8
- 238000005070 sampling Methods 0.000 claims description 14
- 230000009977 dual effect Effects 0.000 claims description 9
- 230000008859 change Effects 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 4
- 238000012937 correction Methods 0.000 claims description 2
- 238000013461 design Methods 0.000 claims description 2
- 230000000873 masking effect Effects 0.000 claims 1
- 238000002595 magnetic resonance imaging Methods 0.000 abstract description 6
- 238000013135 deep learning Methods 0.000 description 6
- 238000012549 training Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 238000003384 imaging method Methods 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000005481 NMR spectroscopy Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000005865 ionizing radiation Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000035479 physiological effects, processes and functions Effects 0.000 description 1
- 238000011158 quantitative evaluation Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/30—Assessment of water resources
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The invention discloses a magnetic resonance image reconstruction method based on a double-domain transducer, which comprises the following steps: carrying out undersampling mask processing on the fully sampled K space data to obtain undersampled K space data; taking the undersampled K space data as the input of a frequency domain network to obtain reconstructed K space data; correcting the reconstructed K space data through a data consistency layer, and obtaining an initial reconstructed image through inverse Fourier transform; and inputting the initial reconstructed image into an image domain network to obtain a final magnetic resonance reconstructed image. The invention accelerates the magnetic resonance imaging speed and obtains high-quality reconstructed images.
Description
Technical Field
The invention belongs to the technical field of image processing, and particularly relates to a magnetic resonance image reconstruction method based on a double-domain transducer.
Background
Magnetic resonance imaging (Magnetic Resonance Imaging, MRI) is currently a clinically used auxiliary examination method with the advantages of non-invasiveness, no ionizing radiation, multiple parameters, etc. However, since raw data of magnetic resonance is acquired sequentially in k-space (k-space), the traversing speed of k-space is limited by physiology and hardware, making it slower than other image examinations. The longer scanning time not only limits the utilization efficiency of the magnetic resonance equipment, but also is easily affected by the artifacts caused by slight movement of the patient in the scanning process, thus preventing the further popularization and development of the magnetic resonance imaging technology.
For magnetic resonance reconstruction, in order to improve imaging speed and obtain high-quality magnetic resonance images, a series of research results are generated, and are mainly divided into two types at present, one is a sparse prior-based method, and the other is a Deep Learning-based method. The former magnetic resonance image reconstruction algorithm includes partial fourier transform, parallel Imaging (Parallel Imaging), compressed sensing (Compressed Sensing, CS), low-rank matrix completion, flow pattern learning, and the like. The deep learning-based method is further divided into a data-driven-based algorithm and a model-driven-based algorithm. The data-driven algorithm establishes a mapping relation between undersampling input and full sampling output through a deep learning algorithm by constructing an end-to-end network model, and can directly obtain a high-quality reconstructed image. The method has the advantages that the operation speed in the prediction stage is in the millisecond level, the method is more in line with the real-time performance in comparison with multiple iterations of a compressed sensing algorithm, and the super-parameters which need to be manually adjusted are far smaller than those of the traditional reconstruction method, but the method has the defect of poor interpretation of the model. The key iteration steps of the traditional algorithm such as compressed sensing and the like are replaced by deep learning network optimization based on the model-driven algorithm, and the method has strong interpretability, but the calculated amount is still large and does not meet the requirement of dynamic magnetic resonance instantaneity.
Most magnetic resonance reconstruction studies heretofore construct a deep-learning reconstruction model only in the image domain, failing to adequately account for frequency domain information. Meanwhile, the cross-domain reconstruction method for optimizing in the frequency domain and the image domain generally obtains magnetic resonance reconstruction results superior to a single-domain model, such as XPDNet, KIKI-Net, hybrid-Cascade-Net and the like which are recently proposed. These convolutional neural network-based models are subject to locality specific to convolution, and it is difficult to construct long-range dependencies of images, resulting in loss of detail and texture in the reconstructed images. In particular, reconstruction algorithms require artifact removal, while aliasing artifacts are typically global, and convolutional networks have difficulty effectively removing artifacts, thus limiting the performance and application of the correlation methods.
Disclosure of Invention
In order to solve the technical problems, the invention provides a magnetic resonance image reconstruction method based on a double-domain transducer, which accelerates the magnetic resonance imaging speed and obtains high-quality reconstructed images.
In order to achieve the above object, the present invention provides a magnetic resonance image reconstruction method based on a dual-domain transducer, comprising:
carrying out undersampling mask processing on the fully sampled K space data to obtain undersampled K space data;
taking the undersampled K space data as the input of a frequency domain network to obtain reconstructed K space data;
correcting the reconstructed K space data through a data consistency layer, and obtaining an initial reconstructed image through inverse Fourier transform;
and inputting the initial reconstructed image into an image domain network to obtain a final magnetic resonance reconstructed image.
Optionally, performing undersampling mask processing on the fully sampled K-space data, and obtaining undersampled K-space data includes:
and acquiring fully sampled K space data through magnetic resonance equipment, and performing undersampling mask processing on the fully sampled K space data by adopting a two-dimensional Gaussian mask according to a preset sampling rate to acquire undersampled K space data.
Optionally, the frequency domain network adopts a U-shaped architecture design, and comprises an encoder, a bottleneck layer, a decoder and jump connection;
the encoder comprises three Swin Transformer blocks, each Swin Transformer block consists of 2 Swin Transformer layers, and the bottom consists of 2 Swin Transformer layers;
the bottleneck layer comprises 2 continuous Swin converterlers layers;
the decoder comprises three continuous Swin transform blocks and a Patch expansion layer;
the skip connection is used to splice the multi-scale features from the encoder with the upsampled features.
Optionally, taking the undersampled K-space data as an input to a frequency domain network, obtaining reconstructed K-space data includes:
dividing the input undersampled K-space data into a plurality of image blocks of size 4*4 that do not overlap each other;
inputting the image block into a linear Patch Embedding layer to convert the image into an ebedding sequence to generate a token sequence;
sequentially inputting the token sequence into three Swin transform blocks and a Patch Merging layer to extract features, downsample and change the size, and obtaining a feature map;
upsampling the feature map by the decoder and adjusting the feature map size to the previous 2 times;
the feature map is restored to the original resolution by the Patch expansion layer and input to the final Linear Projection to obtain reconstructed K-space data.
Optionally, the Swin Transformer layer is composed of a layer normalization module, a window multi-head self-attention module/a sliding window multi-head self-attention module, a multi-layer perceptron layer and residual connection.
Optionally, correcting the reconstructed K-space data through a data consistency layer includes:
and selecting fully sampled K space data in the undersampled K space data, setting the position of a sampling matrix to be 1, keeping the reconstructed K space data consistent with the data of the position corresponding to the fully sampled K space data, and finishing the correction of the reconstructed K space data.
Optionally, inputting the initial reconstructed image into an image domain network, and obtaining a final magnetic resonance reconstructed image includes:
dividing the input initial reconstructed image data into a plurality of image blocks which are 4*4 in size and do not overlap each other;
inputting the image block into a linear Patch Embedding layer to convert the image into an ebedding sequence to generate a token sequence;
sequentially inputting the token sequence into three Swin transform blocks and a Patch Merging layer to extract features, downsample and change the size, and obtaining a feature map;
upsampling the feature map by a decoder and adjusting the feature map size to the previous 2 times;
the feature map is restored to the original resolution by the Patch Expanding layer and input to the final Linear Projection to obtain the final magnetic resonance reconstructed image.
The invention has the technical effects that: the invention discloses a magnetic resonance image reconstruction method based on a double-domain transducer, which provides a magnetic resonance reconstruction model with cross domain network fused with frequency domain and image domain priori knowledge, and can fully utilize the information of the frequency domain and the image domain, wherein each domain consists of a U-shaped encoder decoder structure combined with a Swin transducer, and the two domains are connected through inverse Fourier transform; the method is characterized by effectively reconstructing a magnetic resonance image and simultaneously reducing the influence of aliasing artifacts on the image quality; the network removes the convolution block of the traditional U-Net in the encoder and decoder structure of the double domains, and adopts a Swin transform block as a basic module; the global self-attention mechanism based on the transducer overcomes the defect of limited convolutional operation receptive field, and texture and detail are effectively reserved in the reconstruction process, so that a high-quality reconstructed image is obtained.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application, illustrate and explain the application and are not to be construed as limiting the application. In the drawings:
FIG. 1 is a flow chart of a dual domain transducer-based magnetic resonance image reconstruction method according to an embodiment of the present invention;
FIG. 2 is a diagram of a magnetic resonance image reconstruction network based on a dual domain transducer network in accordance with an embodiment of the present invention;
FIG. 3 is a diagram of a Swin transducer module in a reconstruction network according to an embodiment of the present invention;
FIG. 4 is a graph showing image performance contrast of four reconstruction algorithms of a two-dimensional Gaussian sampling mask at three sampling rates according to an embodiment of the invention;
fig. 5 is a visual contrast diagram of reconstructed images of four reconstruction algorithms under a two-dimensional gaussian sampling mask according to an embodiment of the present invention.
Detailed Description
It should be noted that, in the case of no conflict, the embodiments and features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the accompanying drawings in conjunction with embodiments.
It should be noted that the steps illustrated in the flowcharts of the figures may be performed in a computer system such as a set of computer executable instructions, and that although a logical order is illustrated in the flowcharts, in some cases the steps illustrated or described may be performed in an order other than that illustrated herein.
As shown in fig. 1, the present embodiment provides a magnetic resonance image reconstruction method based on a dual-domain transducer, which includes:
step 1: firstly, undersampling mask processing is carried out on fully sampled K space data to obtain undersampled K space data;
step 2: then taking the undersampled K space data as the input of a frequency domain network, and reconstructing the K space data;
step 3: correcting the reconstructed K space data by using a data consistency layer, and obtaining an initial reconstructed image through inverse Fourier transform;
step 4: and finally, sending the initial reconstructed image to an image domain network to obtain a final magnetic resonance reconstructed image.
The frequency domain network can fill the missing undersampled K space data, so that the undersampled K space data is richer, and more information is provided for the subsequent image domain network.
The data consistency layer is used as a common strategy for improving the reconstruction quality, and the reconstruction data with the sampling matrix position of 1 is kept consistent with the data at the position corresponding to the original data by utilizing the numerical value of part of undersampled data belonging to the original data, so that the aim of correcting the predicted data is fulfilled.
And step 1, acquiring magnetic resonance full-sampling data by using magnetic resonance equipment, wherein an acquisition format comprises original K space data, and generating a corresponding full-sampling reconstructed image through inverse Fourier transform. And undersampling is carried out on the original data at 10%,20% and 30% of sampling rate by adopting a two-dimensional Gaussian mask, so that undersampled K space data are obtained.
The frequency domain network in the step 2 is designed based on a U-shaped architecture and consists of an encoder, a bottleneck layer, a decoder and jump connection. The encoder consisted of three Swin fransformer blocks, each Swin fransformer block consisting of 2 Swin fransformer layers and the bottom consisting of 2 Swin fransformer layers. The input original image is firstly divided into a plurality of image blocks which are 4*4 in size and are not overlapped with each other, and a linear Patch editing layer is input to convert the image into an ebedding sequence. The resulting token sequence is then sequentially input into three Swin transform blocks and a Patch Merry layer, where the Patch Merry layer is used to downsample and resize, the Swin transform blocks are used to extract features. The symmetric decoder consists of three consecutive Swin transform blocks and a Patch expansion layer, where the Patch expansion layer is used for upsampling and resizing the feature map by a factor of 2 before. The last Patch expansion layer restores the feature map to the original resolution and inputs to the final Linear Projection to get the reconstructed K-space data. The bottleneck layer comprises 2 continuous Swin transducer layers; the decoder comprises three continuous Swin transform blocks and a Patch Expanding layer; the jump connection is used for splicing the multi-scale features from the encoder with the upsampled features; the stitching operation will stitch together the plurality of tensors along a given axis along the direction of that axis. The data input to the network is frequency domain data, so that the data is called a frequency domain network, and the frequency domain data is converted into image domain data by inverse fourier transform, so that the data is called an image domain network, and the image domain network has the same structure as the frequency domain network.
Compared with the traditional transducer, the Swin transducer adopted in the network balances the calculation amount of the multi-scale global information and the model by designing a sliding window and combining the advantages of the convolutional neural network. Has the following advantages: (1) By limiting the use of self-attention within the window, greater efficiency is achieved. (2) Through the movement, interaction is realized between two adjacent windows, and cross-window connection is also realized between the upper layer and the lower layer, so that the phase change achieves a global modeling effect. (3) The hierarchical structure is not only very flexible to model the information of individual scales but also the computational complexity grows linearly with the image size. The Swin transducer layer is composed of a layer normalization module, a window multi-head self-attention module/a sliding window multi-head self-attention module, a multi-layer perceptron layer and residual error connection.
And 3, correcting the reconstructed K space data by using a data consistency layer, wherein the data consistency layer is used as a common strategy for improving the reconstruction quality, and the reconstructed data with the sampling matrix position of 1 is kept consistent with the data at the position corresponding to the original data by utilizing the numerical value of part of undersampled data belonging to the original data, so that the aim of correcting the predicted data is fulfilled. The data consistency layer operates as follows:
wherein F is K,n Is the output of the frequency domain network in K-space, y is the undersampled K-space data, Ω represents the frequency domain sampled spatial data, j represents a point in space, λ is a hyper-parameter.
The image domain network in the step 4 is consistent with the frequency domain network structure in the step 2. The overall loss function of the magnetic resonance reconstruction method based on the double-domain transducer network consists of a weighted sum of frequency domain loss and image domain loss, wherein each domain calculates normalized root mean square error (Normalized Root Mean Square Error, NRMSE) of the domain respectively and is used for constraining information of different domains, and the specific formula is as follows:
wherein F is i And f i The full sample K-space data and the image domain data representing the ith sample in the training set,and->The reconstructed K-space data and image domain data representing the ith sample in the training set, respectively, N being the total number of samples in the training set. To ensure that greater weight is applied to the final network output, ω takes a value of 0.001.
Correcting the reconstructed K space data through a data consistency layer, and obtaining an initial reconstructed image through inverse Fourier transform; and inputting the initial reconstructed image into an image domain network to obtain a final magnetic resonance reconstructed image.
The model training adopts CPU Intel Core i9-10920X 3.50GHz,GeForce RTX 2080Ti GPU (11 GB), windows 10 operating system, tensorflow 2.4 deep learning framework, and the language environment is Python 3.8. The network model optimizer was Adam, the initial learning rate was set to 0.001, the batch throughput was set to 8, and the number of model iterations was set to 200. When the training times reach 20 times without further reduction of loss, model training is finished in advance and the best model is saved.
Fig. 2 is a diagram showing a magnetic resonance image reconstruction network structure based on a dual domain transducer network according to an embodiment of the present invention; providing a magnetic resonance reconstruction model with cross domain network fused with frequency domain and image domain priori knowledge, and fully utilizing the information of the frequency domain and the image domain, wherein each domain consists of a U-shaped encoder and decoder structure combined with a Swin Transformer, and the two domains are connected through inverse Fourier transform;
fig. 3 is a diagram showing a Swin transducer module in a reconstruction network according to an embodiment of the present invention; the Swin transducer module is composed of a layer normalization module, a window multi-head self-attention module/a sliding window multi-head self-attention module, a multi-layer perceptron layer and residual connection.
In the reconstruction, in addition to the reconstruction network proposed by the present invention, the previously proposed reconstruction networks KIKI-Net, XPDNet and MDReconNet with excellent performance are used for comparison. For comprehensive evaluation, the effectiveness of the network is proposed, and the reconstruction results of the magnetic resonance image under the two-dimensional Gaussian masks with the sampling rates of 10%,20% and 30% are compared. Three quantitative evaluation indices, including peak signal-to-noise ratio (Peak Signal to Noise Ratio, PSNR), normalized root mean square error (Normalized Root Mean Square Error, NRMSE) and structural similarity (Structure Similarity Index Measure, SSIM), were used to quantitatively evaluate the reconstructed image quality. The reconstruction results of the four different networks are shown in fig. 4, and the nuclear magnetic resonance image reconstruction network based on the dual-domain converter network has better performance than other networks, and has consistent results under different evaluation indexes and different undersampling modes. Fig. 5 is a visual comparison, and compared with other algorithms, the network provided by the invention can better remove artifacts caused by undersampling, reconstruct high-frequency details, obtain more prominent texture structures and reconstruct clear images.
The foregoing is merely a preferred embodiment of the present application, but the scope of the present application is not limited thereto, and any changes or substitutions easily conceivable by those skilled in the art within the technical scope of the present application should be covered in the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (7)
1. The magnetic resonance image reconstruction method based on the double-domain transducer is characterized by comprising the following steps of:
carrying out undersampling mask processing on the fully sampled K space data to obtain undersampled K space data;
taking the undersampled K space data as the input of a frequency domain network to obtain reconstructed K space data;
correcting the reconstructed K space data through a data consistency layer, and obtaining an initial reconstructed image through inverse Fourier transform;
and inputting the initial reconstructed image into an image domain network to obtain a final magnetic resonance reconstructed image.
2. The dual domain fransformer-based magnetic resonance image reconstruction method of claim 1, wherein undersampling masking the fully sampled K-space data to obtain undersampled K-space data comprises:
and acquiring fully sampled K space data through magnetic resonance equipment, and performing undersampling mask processing on the fully sampled K space data by adopting a two-dimensional Gaussian mask according to a preset sampling rate to acquire undersampled K space data.
3. The method for reconstructing a magnetic resonance image based on a dual domain transducer according to claim 1, wherein the frequency domain network adopts a U-shaped architecture design, including an encoder, a bottleneck layer, a decoder and a jump connection;
the encoder comprises three Swin Transformer blocks, each Swin Transformer block consists of 2 Swin Transformer layers, and the bottom consists of 2 Swin Transformer layers;
the bottleneck layer comprises 2 continuous Swin converterlers layers;
the decoder comprises three continuous Swin transform blocks and a Patch expansion layer;
the skip connection is used to splice the multi-scale features from the encoder with the upsampled features.
4. The dual domain fransformer-based magnetic resonance image reconstruction method as set forth in claim 3, wherein taking the undersampled K-space data as input to a frequency domain network, obtaining reconstructed K-space data comprises:
dividing the input undersampled K-space data into a plurality of image blocks of size 4*4 that do not overlap each other;
inputting the image block into a linear Patch Embedding layer to convert the image into an ebedding sequence to generate a token sequence;
sequentially inputting the token sequence into three Swin transform blocks and a Patch Merging layer to extract features, downsample and change the size, and obtaining a feature map;
upsampling the feature map by the decoder and adjusting the feature map size to the previous 2 times;
and restoring the feature map to the original resolution through a Patch Expanding layer, and inputting the feature map to a final linearProjecting to obtain reconstructed K space data.
5. The method for reconstructing a magnetic resonance image based on a dual domain transducer of claim 3, wherein the Swin transducer layer is composed of a layer normalization module, a window multi-head self-attention module/sliding window multi-head self-attention module, a multi-layer perceptron layer and a residual connection.
6. The dual domain fransformer-based magnetic resonance image reconstruction method of claim 1, wherein correcting the reconstructed K-space data through a data consistency layer comprises:
and selecting fully sampled K space data in the undersampled K space data, setting the position of a sampling matrix to be 1, keeping the reconstructed K space data consistent with the data of the position corresponding to the fully sampled K space data, and finishing the correction of the reconstructed K space data.
7. The dual domain transducer based magnetic resonance image reconstruction method as set forth in claim 1, wherein inputting the initial reconstructed image into an image domain network, obtaining a final magnetic resonance reconstructed image comprises:
dividing the input initial reconstructed image data into a plurality of image blocks which are 4*4 in size and do not overlap each other;
inputting the image block into a linear Patch Embedding layer to convert the image into an ebedding sequence to generate a token sequence;
sequentially inputting the token sequence into three Swin transform blocks and a Patch Merging layer to extract features, downsample and change the size, and obtaining a feature map;
upsampling the feature map by a decoder and adjusting the feature map size to the previous 2 times;
and restoring the characteristic image to the original resolution through a Patch Expanding layer, and inputting the characteristic image to a final linearProjecting to obtain a final magnetic resonance reconstruction image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311586159.1A CN117576240A (en) | 2023-11-24 | 2023-11-24 | Magnetic resonance image reconstruction method based on double-domain transducer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311586159.1A CN117576240A (en) | 2023-11-24 | 2023-11-24 | Magnetic resonance image reconstruction method based on double-domain transducer |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117576240A true CN117576240A (en) | 2024-02-20 |
Family
ID=89863937
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311586159.1A Pending CN117576240A (en) | 2023-11-24 | 2023-11-24 | Magnetic resonance image reconstruction method based on double-domain transducer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117576240A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117890844A (en) * | 2024-03-15 | 2024-04-16 | 山东大学 | Magnetic resonance image reconstruction method based on optimized mask model |
-
2023
- 2023-11-24 CN CN202311586159.1A patent/CN117576240A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117890844A (en) * | 2024-03-15 | 2024-04-16 | 山东大学 | Magnetic resonance image reconstruction method based on optimized mask model |
| CN117890844B (en) * | 2024-03-15 | 2024-05-24 | 山东大学 | Magnetic resonance image reconstruction method based on optimized mask model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Deep learning for fast MR imaging: A review for learning reconstruction from incomplete k-space data | |
| Montalt-Tordera et al. | Machine learning in magnetic resonance imaging: image reconstruction | |
| Zhao et al. | SwinGAN: A dual-domain Swin Transformer-based generative adversarial network for MRI reconstruction | |
| Bao et al. | Undersampled MR image reconstruction using an enhanced recursive residual network | |
| CN103472419A (en) | Magnetic-resonance fast imaging method and system thereof | |
| CN109214989A (en) | Single image super resolution ratio reconstruction method based on Orientation Features prediction priori | |
| Lin et al. | Deep learning for low-field to high-field MR: image quality transfer with probabilistic decimation simulator | |
| Aghabiglou et al. | Projection-Based cascaded U-Net model for MR image reconstruction | |
| Zou et al. | Multi-scale deformable transformer for multi-contrast knee MRI super-resolution | |
| CN117333750A (en) | Spatial registration and local global multi-scale multi-modal medical image fusion method | |
| CN117576240A (en) | Magnetic resonance image reconstruction method based on double-domain transducer | |
| Cui et al. | Motion artifact reduction for magnetic resonance imaging with deep learning and k-space analysis | |
| Ekanayake et al. | Multi-branch Cascaded Swin Transformers with Attention to k-space Sampling Pattern for Accelerated MRI Reconstruction | |
| Yang et al. | MGDUN: An interpretable network for multi-contrast MRI image super-resolution reconstruction | |
| Lim et al. | Motion artifact correction in fetal MRI based on a Generative Adversarial network method | |
| Qu et al. | Radial magnetic resonance image reconstruction with a deep unrolled projected fast iterative soft-thresholding network | |
| CN117974468A (en) | Multi-mode medical image fusion method for global and local feature interaction parallelism | |
| CN112669400B (en) | Dynamic MR reconstruction method based on deep learning prediction and residual error framework | |
| CN116758120A (en) | 3T MRA-7T MRA prediction method based on deep learning | |
| CN116977458A (en) | Method and system for generating multichannel space support diagram | |
| CN113866694B (en) | Rapid three-dimensional magnetic resonance T1 quantitative imaging method, system and medium | |
| Ji et al. | Deep learning-based magnetic resonance image super-resolution: a survey | |
| CN116705251A (en) | Brain MR image multidirectional normalization system based on deep learning | |
| CN117333571B (en) | Reconstruction method, system, equipment and medium of magnetic resonance image | |
| Xu et al. | Multi-modal brain MRI images enhancement based on framelet and local weights super-resolution |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |