CN117011409A - Multi-position physical intelligent high-definition diffusion magnetic resonance data generation method - Google Patents
Multi-position physical intelligent high-definition diffusion magnetic resonance data generation method Download PDFInfo
- Publication number
- CN117011409A CN117011409A CN202311004982.7A CN202311004982A CN117011409A CN 117011409 A CN117011409 A CN 117011409A CN 202311004982 A CN202311004982 A CN 202311004982A CN 117011409 A CN117011409 A CN 117011409A
- Authority
- CN
- China
- Prior art keywords
- diffusion
- data
- magnetic resonance
- definition
- motion
- 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.)
- Granted
Links
- 238000009792 diffusion process Methods 0.000 title claims abstract description 259
- 238000000034 method Methods 0.000 title claims abstract description 32
- 238000012549 training Methods 0.000 claims abstract description 101
- 230000033001 locomotion Effects 0.000 claims abstract description 86
- 238000004088 simulation Methods 0.000 claims abstract description 53
- 238000005070 sampling Methods 0.000 claims abstract description 39
- 230000005284 excitation Effects 0.000 claims abstract description 25
- 238000010586 diagram Methods 0.000 claims abstract description 21
- 238000010587 phase diagram Methods 0.000 claims abstract description 19
- 230000035945 sensitivity Effects 0.000 claims abstract description 16
- 238000013528 artificial neural network Methods 0.000 claims abstract description 12
- 230000009466 transformation Effects 0.000 claims abstract description 6
- 230000001131 transforming effect Effects 0.000 claims abstract description 6
- 238000003384 imaging method Methods 0.000 claims description 26
- 239000011159 matrix material Substances 0.000 claims description 16
- 210000004556 brain Anatomy 0.000 claims description 8
- 230000014509 gene expression Effects 0.000 claims description 4
- 238000002788 crimping Methods 0.000 claims description 3
- 210000001519 tissue Anatomy 0.000 claims description 3
- 230000003187 abdominal effect Effects 0.000 claims description 2
- 210000004185 liver Anatomy 0.000 claims description 2
- 210000000496 pancreas Anatomy 0.000 claims description 2
- 210000002307 prostate Anatomy 0.000 claims description 2
- 230000011218 segmentation Effects 0.000 claims description 2
- 238000005259 measurement Methods 0.000 abstract description 6
- 238000002597 diffusion-weighted imaging Methods 0.000 description 8
- 230000004913 activation Effects 0.000 description 6
- 238000013527 convolutional neural network Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 210000001015 abdomen Anatomy 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000002592 echocardiography Methods 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 210000004197 pelvis Anatomy 0.000 description 2
- BZLVMXJERCGZMT-UHFFFAOYSA-N Methyl tert-butyl ether Chemical compound COC(C)(C)C BZLVMXJERCGZMT-UHFFFAOYSA-N 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 229940079593 drug Drugs 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000002595 magnetic resonance imaging Methods 0.000 description 1
- 230000004879 molecular function Effects 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R33/00—Arrangements or instruments for measuring magnetic variables
- G01R33/20—Arrangements or instruments for measuring magnetic variables involving magnetic resonance
- G01R33/44—Arrangements or instruments for measuring magnetic variables involving magnetic resonance using nuclear magnetic resonance [NMR]
- G01R33/48—NMR imaging systems
- G01R33/54—Signal processing systems, e.g. using pulse sequences ; Generation or control of pulse sequences; Operator console
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- 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
-
- 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
- G06N3/08—Learning methods
-
- 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
- G06T11/005—Specific pre-processing for tomographic reconstruction, e.g. calibration, source positioning, rebinning, scatter correction, retrospective gating
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10088—Magnetic resonance imaging [MRI]
-
- 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)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Software Systems (AREA)
- Pure & Applied Mathematics (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Algebra (AREA)
- Signal Processing (AREA)
- High Energy & Nuclear Physics (AREA)
- Condensed Matter Physics & Semiconductors (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The multi-part physical intelligent high-definition diffusion magnetic resonance data generation method comprises 1) generating a simulated high-definition diffusion magnetic resonance amplitude diagram by using a diffusion physical model; 2) Generating a high-order simulation motion phase of each excitation diffusion image based on the physical motion model; 3) Combining the simulated diffusion magnetic resonance amplitude diagram, the high-order simulated motion phase diagram and the acquired multichannel coil sensitivity map, transforming to k space by utilizing Fourier transformation to obtain fully sampled network training tag data, and 4) adding Gaussian noise to obtain noisy k space data; 5) Undersampling noisy k-space data according to a sampling track of an acquisition sequence of high-definition diffusion data of a target to obtain simulated multichannel multi-excitation diffusion magnetic resonance data with motion artifact interference; 6) Repeating 1) to 5) to generate network training input data and full-sampling training label data; 7) And (3) utilizing the generated diffusion magnetic resonance data set training neural network to realize reconstruction or denoising of the actual measurement high-definition diffusion magnetic resonance data.
Description
Technical Field
The invention relates to a generation method based on multichannel multi-excitation diffusion image data, in particular to a generation method for generating paired multi-part physical intelligent high-definition diffusion magnetic resonance data by using a diffusion physical model and a high-order physical motion phase model, wherein the generated data can be used for training of a reconstructed neural network.
Background
Diffusion weighted imaging (Diffusion weighted imaging) is a way to assess human molecular function and microstructure and to detect the diffusion movement of water molecules in tissues without intrusion (v.baliyan et al., "Diffusion weighted imaging: technique and applications," World Journal of Radiology,8,785,2016). Multi-shot planar echo imaging techniques have the ability to improve resolution and reduce low-distortion in diffusion weighting applications (H.An, X.Ma, Z.Pan, H.Guo, E.Y.P.Lee, "Qualitative and quantitative comparison of image quality between single-shot echo-planar and interleaved multi-shot echo-planar dispersion-weighted imaging in female pelvis," European radiology,30,1876-1884,2020). However, there is a severe phase error between the different excitations, resulting in severe motion artifacts (A.W.Anderson, J.C.Gore, "Analysis and correction of motion artifacts in diffusion weighted imaging," Magnetic Resonance in Medicine,32,379-387,1994).
Recently, the deep learning approach has shown great potential in multi-shot diffusion weighted imaging (Aggarwal.H.K., M.Mani, M.Jacob, "MoDL-MUSSELS: model-based deep learning for multi-shot sensitivity-encoded diffusion MRI", IEEE Transactions on Medical Imaging,39,1268-1277,2019). However, the multi-shot diffusion weighted image lacks high quality training labels, the training labels generated using conventional iterative reconstruction methods (C.Qian et al., "A paired phase and magnitude reconstruction for advanced diffusion-weighted imaging," IEEE Transactions on Biomedical Engineering, DOI:10.1109/TBME.2023.3288031, 2023) are a popular solution (F.Wang et al., "Multiple b-value model-based residual network (MORN) for accelerated high-resolution-weighted imaging," IEEE Journal ofBiomedical and Health Informatics,26,4575-4586,2022), but it greatly limits the potential of these intelligent reconstruction methods: (1) The traditional iterative reconstruction method generally needs extremely long time to generate a large number of training labels, and the parameter adjustment is complicated, so that the robust reconstruction is difficult to obtain on a large actual measurement data set; (2) The performance of the deep learning method will be limited by the tag generation method; (3) Due to insufficient data sets, the deep learning method is poor in generalization under the reconstruction of multiple scenes such as multiple resolutions, multiple b values, multiple manufacturers, multiple undersamples and the like; (4) The measured data collection is time consuming and is more limited because of privacy concerns. Meanwhile, the traditional iterative reconstruction method is only suitable for rigid motion, and the abdomen has more complex elastic motion, so that the invention adopts a high-order physical motion phase simulation model in motion phase simulation.
In summary, the current intelligent reconstruction method of high-definition diffusion imaging is limited by the bottleneck of high-quality paired training data, and the potential of the intelligent reconstruction method cannot be effectively exerted.
Disclosure of Invention
The invention aims to provide a multi-position physical intelligent high-definition diffusion magnetic resonance data generation method capable of rapidly generating multi-b-value, multi-resolution, multi-sampling-rate and multi-sequence high-definition diffusion imaging data.
The invention comprises the following steps:
1) Generating a simulation high-definition diffusion magnetic resonance amplitude diagram by using diffusion physical models such as diffusion indexes, tensors, kurtosis and the like;
2) Generating a high-order simulation motion phase diagram of each excitation diffusion image based on the physical motion model;
3) Combining the generated simulation high-definition diffusion magnetic resonance amplitude diagram, the high-order simulation motion phase diagram and the acquired multichannel coil sensitivity map, and transforming the simulation high-definition diffusion magnetic resonance amplitude diagram, the high-order simulation motion phase diagram and the acquired multichannel coil sensitivity map into k space by utilizing Fourier transformation to obtain fully-sampled network training label data;
4) Adding Gaussian noise into the full-sampling training label data to obtain noisy k-space data;
5) Undersampling noisy k-space data according to a sampling track of an acquisition sequence of high-definition diffusion data of a target to obtain simulated multichannel multi-excitation diffusion magnetic resonance data with motion artifact interference, wherein the simulated multichannel multi-excitation diffusion magnetic resonance data is used as input data of network training;
6) Repeating steps 1) to 5) in a large number to generate a large number of paired network training input data and full-sampling training label data;
7) Training a neural network by using the generated diffusion magnetic resonance data set, wherein the network is used for reconstructing or denoising the high-definition diffusion magnetic resonance data after training.
In step 1), the generated simulated high-definition diffusion magnetic resonance amplitude graph can be generated by diffusion physical models such as diffusion indexes, tensors, kurtosis and the like:
(diffusion single exponential model) m (b) =exp { -bdadc } m 0 (1)
Wherein,for diffusion magnitude image with diffusion b value b, < ->For diffusion images where no diffusion gradient is applied, the ADC is an apparent diffusion coefficient matrix.
(diffusion tensor model) m (b, g) =exp { -bg T Dg}m 0 (2)
Wherein,for diffusion amplitude image with diffusion b value b in diffusion gradient direction g, +.>For a diffusion image to which no diffusion gradient is applied, g represents a single diffusion gradient directionBit vector, T is the transposed symbol, and D is the diffusion tensor matrix.
(diffusion kurtosis model)Wherein (1)>For diffusion amplitude image with diffusion b value b in diffusion gradient direction g, +.>For a diffusion image to which a diffusion gradient is not applied, g represents a unit vector of a diffusion gradient direction, D is a diffusion tensor matrix, and K is a diffusion kurtosis matrix.
The constants of ADC, D, K and the like with tissue diffusion information in the diffusion single-index, tensor, kurtosis and other physical models can be calculated by acquiring actual diffusion images with enough diffusion b values and diffusion direction numbers through linear fitting and other algorithms.
Thus, given the diffusion direction g and b value b, a diffusion magnitude image m of the target diffusion direction and b value can be generated with constants of ADC, D, K, etc. known.
In step 2), the high-order simulated motion phases of each excitation-diffusion image can be generated by a physical motion model:
wherein the diagonal matrixThe phase obtained by simulation of the model is that N and M are the length of the frequency code and the phase code of the image respectively, x and y are coordinates on a selected layer of the two-dimensional image, and i is an imaginary symbol. L is the highest order of the single expressions in the polynomial model, m and L-m are the powers of the single expressions x and y in the current first order polynomial, A lm Is the corresponding item x m y l-m Is a coefficient of (a).By using an L-order polynomial phase model, J motion phases can be obtained by fitting the motion phases of the measured data through phase unwrapping and phase crimping operations>J is the excitation times of the multi-excitation diffusion weighted data to be reconstructed; and then according to the distribution rule of polynomial coefficients in the L-order polynomial phase model, a large number of polynomial coefficients meeting the distribution rule can be obtained to form a simulation phase set. The motion conditions of different parts are different, the brain, the prostate and other parts can be approximately modeled as rigid motion, and a motion model with L=3 is used for simulating motion phases; the abdominal liver and pancreas have more complex elastic motions, and the motion model of l=7 or higher order is used to approximate the simulated motion phase.
In step 3), the generated simulated high definition diffusion magnetic resonance amplitude mapHigh-order simulation motion phase diagram->And an acquired multichannel coil sensitivity map +.>Combining according to the following formula to obtain fully sampled network training label data:
wherein,is +.>Or in a non-Cartesian coordinate system>Fourier transform operator of>Is synthesized H channel J-time excited full-sampling k-space data as network training label data.
In step 4), gaussian noise is added to the fully sampled training tag data to obtain noisy k-space data as follows:
wherein,is Gaussian noise with real and imaginary parts satisfying mean value mu and variance sigma, < >>Is noisy fully sampled k-space data with a signal-to-noise ratio (dB) of X GT And calculating to obtain the noiseless reference diagram.
In step 5), undersampling the noisy k-space data according to a sampling track of an acquisition sequence of the high-definition diffusion data of the target to obtain simulated multichannel multi-excitation diffusion magnetic resonance data with motion artifact interference, wherein the simulated multichannel multi-excitation diffusion magnetic resonance data is used as input data of network training and comprises the following steps:
wherein,is simulated multichannel multi-excitation diffusion magnetic resonance data of motion artifact interference, is taken as input data of network training, and is a sampling template ∈>Can be used forSampling tracks for a plurality of multi-excitation diffusion sequences, in the case of Cartesian coordinates +.>Sampling template->Multi-shot staggered planar echo sequence (multi-shot interleaved echo planer imaging, ms-iEPI), multi-shot read-out-segmented echo planer imaging, ms-rsEPI), etc. can be selected; when in a non-Cartesian coordinate systemSampling template->A multi-shot helical (ms-helical) sampling sequence, a multi-shot propeller (ms-propeller) sampling sequence, and the like may be selected. The number of shots was set to J.
In step 6), repeating steps 1) to 5) K times to generate K-component-pair network training input data X inp And full sample training tag data X GT Thereby constructing a high-quality paired high-definition diffusion imaging training data set based on physical intelligence and containing K samples.
In step 7), the generated diffusion magnetic resonance data set may be used to train a neural network, after which the network may be used to reconstruct or denoise the high-definition diffusion magnetic resonance data.
The invention provides a physical intelligent data generation method based on a physical intelligent driving data generation scheme (Q.Yang, Z.Wang, K.Guo, C.Cai and X.Qu, "Physics-driven syntheticdata learning for biomedical magnetic resonance: the imaging Physics based data synthesis paradigm for artificial intelligence," IEEE SignalProcessing Magazine,40,129-140,2023). Compared with brain rigid body motion modeling (C.Qian et al, "Physics-informed deep diffusion reconstruction: break the bottleneck of training data in artificial intelligence", arXiv:2210.11388,2023), the random high-order motion phase model provided by the invention is suitable for fitting non-rigid body elastic motion, can rapidly generate multi-b-value, multi-resolution, multi-sampling rate and multi-sequence high-definition diffusion imaging data, breaks through the bottleneck that the existing high-definition diffusion magnetic resonance intelligent network depends on actual measurement data, lacks high-quality training labels, and has wide application prospect.
Drawings
FIG. 1 is a flow chart of an embodiment of the present invention.
FIG. 2 is a sequence sampling template for constructing undersampled network input dataRespectively (a) 4-time excitation staggered plane echo sequences and (b) 4-time excitation read-out dimension segmented plane echo sequences.
FIG. 3 is a representation of physical intelligent simulation data versus image domain. (a) brain high-definition diffusion data based on 4-shot staggered plane echoes, (b) abdomen high-definition diffusion data based on 4-shot segmented plane echoes, and (c) brain diffusion data based on a single-shot plane echo sequence.
Detailed Description
The embodiment of the invention comprises a specific process of generating brain and abdomen multi-excitation physical intelligent high-definition diffusion magnetic resonance data, and the method for generating the multi-position physical intelligent high-definition diffusion magnetic resonance data is described in detail by combining with a drawing.
Example 1
The brain high-definition diffusion data generation and reconstruction network training process based on four excitation staggered plane echoes is shown in the flow chart of fig. 1, and is described in detail as follows:
the embodiment of the invention comprises the following steps:
1) Generating a simulation high-definition diffusion magnetic resonance amplitude diagram by using diffusion physical models such as diffusion indexes, tensors, kurtosis and the like;
2) Generating a high-order simulation motion phase diagram of each excitation diffusion image based on the physical motion model;
3) Combining the generated simulation high-definition diffusion magnetic resonance amplitude diagram, the high-order simulation motion phase diagram and the acquired multichannel coil sensitivity map, and transforming the simulation high-definition diffusion magnetic resonance amplitude diagram, the high-order simulation motion phase diagram and the acquired multichannel coil sensitivity map into k space by utilizing Fourier transformation to obtain fully-sampled network training label data;
4) Adding Gaussian noise into the full-sampling training label data to obtain noisy k-space data;
5) Undersampling noisy k-space data according to a sampling track of an acquisition sequence of high-definition diffusion data of a target to obtain simulated multichannel multi-excitation diffusion magnetic resonance data with motion artifact interference, wherein the simulated multichannel multi-excitation diffusion magnetic resonance data is used as input data of network training;
6) Repeating steps 1) to 5) in a large number to generate a large number of paired network training input data and full-sampling training label data;
7) Training a neural network by using the generated diffusion magnetic resonance data set, wherein the network is used for reconstructing or denoising the high-definition diffusion magnetic resonance data after training.
In step 1), the generated simulated high definition diffusion magnetic resonance amplitude map may be generated by a diffusion tensor physical model:
(diffusion tensor model) m (b, g) =exp { -bg T Dg}m 0 (8)
Wherein,for diffusion amplitude image with diffusion b value b in diffusion gradient direction g, +.>For a diffusion image without applied diffusion gradient, T is a transposed symbol, D is a diffusion tensor matrix, and the diffusion gradient is represented by acquiring 1 b value of 0s/mm 2 M of (2) 0 And 64 diffusion direction numbers b of 1000s/mm 2 The actual acquired diffusion image of (2) is calculated by a linear fitting algorithm.
Thus, given a diffusion direction g of [0, 1 ] given D is known]And b has a value b of 1000s/mm 2 A diffusion amplitude image m (g= [0, 1) of the target diffusion direction and b value can be generated],b＝1000)。
In step 2), the high-order simulated motion phases of each excited diffusion image can be generated by a physical motion model:
wherein the diagonal matrixIs the phase obtained by simulation of the model, the integer x, y E [1,256 ]]Coordinates at the level are selected for the two-dimensional image, i being an imaginary symbol. L=3 is the single highest order in the polynomial model. By using an L-order polynomial phase model, 4 motion phases can be obtained by fitting the motion phases of the measured data through phase unwrapping and phase wrapping operations>The target multi-shot diffusion weighted data to be reconstructed is four shots.
In step 3), the generated simulated diffusion magnetic resonance amplitude map is generatedHigh-order simulation motion phase diagramAnd an acquired multichannel coil sensitivity map +.>Combining according to the following formula to obtain fully sampled network training label data:
wherein,is a two-dimensional Cartesian systemUnder the coordinate system->Or in a non-Cartesian coordinate system>Fourier transform operator of>Is synthesized 8-channel 4-shot fully sampled k-space data as network training tag data.
In step 4), gaussian noise is added to the fully sampled training tag data to obtain noisy k-space data:
wherein,is Gaussian noise with real and imaginary parts meeting the mean value of 0 and the variance of 0.05,is noisy fully sampled k-space data with a signal-to-noise ratio of X GT Calculated for the noiseless reference diagram is 10dB.
In step 5), setting a sampling track of an acquisition sequence of high-definition diffusion data of a target as a four-time excitation staggered plane echo sequence, undersampling noisy k-space data to obtain simulation 8-channel 4-time excitation diffusion magnetic resonance data of motion artifact interference, wherein the simulation 8-channel 4-time excitation diffusion magnetic resonance data is used as input data of network training:
wherein,the simulation multichannel multi-excitation diffusion magnetic resonance data of motion artifact interference are used as input data of network training. Sampling template->The interleaved planar echo sequence (4 shot-iEPI) may be excited four times, as shown in fig. 2 (a).
In step 6), repeating steps 1) to 5) 100000 times, each time setting different parameters g, b, A lm μ, σ. Generating 100000 group pairs of network training input data X inp And full sample training tag data X GT Thereby constructing a physical intelligence based high quality paired high definition diffusion imaging training dataset containing 100000 samples. Fig. 3 (a) illustrates a pair of physically intelligent generated network training tags and input band motion artifact data.
In step 7), the generated high quality paired high definition diffusion imaging training dataset of physical intelligence can be used to train a neural network. A convolutional neural network is designed that contains 5 identical reconstruction modules, each comprising 6 convolutional layers, each comprising 32 convolutional kernels of 3 x 3 size and a layer of linear rectifying activation functions (the sixth layer does not contain a linear rectifying activation function). After training, the convolutional neural network can be used for reconstructing actual measurement 4-time excitation high-definition diffusion imaging data.
The physical intelligent data generation specific scheme can rapidly generate paired high-definition diffusion imaging data with multiple b values and multiple resolutions, breaks through the bottleneck that the existing intelligent high-definition diffusion imaging network depends on actual measured data and is lack of high-quality training labels, and has wide application prospect
Example 2
The abdomen high-definition diffusion data generation and reconstruction network training process based on four-time excitation readout dimension segmentation sampling is shown in the flow chart of fig. 1, and is described in detail as follows:
the embodiment of the invention comprises the following steps:
1) Generating a simulation high-definition diffusion magnetic resonance amplitude diagram by using diffusion physical models such as diffusion indexes, tensors, kurtosis and the like;
2) Generating a high-order simulation motion phase diagram of each excitation diffusion image based on the physical motion model;
3) Combining the generated simulation high-definition diffusion magnetic resonance amplitude diagram, the high-order simulation motion phase diagram and the acquired multichannel coil sensitivity map, and transforming the simulation high-definition diffusion magnetic resonance amplitude diagram, the high-order simulation motion phase diagram and the acquired multichannel coil sensitivity map into k space by utilizing Fourier transformation to obtain fully-sampled network training label data;
4) Adding Gaussian noise into the full-sampling training label data to obtain noisy k-space data;
5) Undersampling noisy k-space data according to a sampling track of an acquisition sequence of high-definition diffusion data of a target to obtain simulated multichannel multi-excitation diffusion magnetic resonance data with motion artifact interference, wherein the simulated multichannel multi-excitation diffusion magnetic resonance data is used as input data of network training;
6) Repeating steps 1) to 5) in a large number to generate a large number of paired network training input data and full-sampling training label data;
7) Training a neural network by using the generated diffusion magnetic resonance data set, wherein the network is used for reconstructing or denoising the high-definition diffusion magnetic resonance data after training.
In step 1), the generated simulated high definition diffusion magnetic resonance amplitude plot may be generated by a diffusion single-index physical model:
(diffusion single exponential model) m (b) =exp { -bdadc } m 0 (13)
Wherein,is a diffusion amplitude image with diffusion b value b, a>For diffusion images without applied diffusion gradients, ADC is an apparent diffusion coefficient matrix obtained by acquiring 1 b value of 0s/mm 2 M of (2) 0 And 3 diffusion directions b have a value of 1000s/mm 2 The actual acquired diffusion image of (2) is calculated by a linear fitting algorithm to obtain the ADC.
Thus, given a spread b value b of 1000s/mm, given that the ADC is known 2 A diffusion amplitude map of the target diffusion b value can be generatedImage m (b=1000).
In step 2), the high-order simulated motion phases of each excited diffusion image can be generated by a physical motion model:
wherein the diagonal matrixIs the phase obtained by simulation of the model, the integer x, y E [1,256 ]]Coordinates on a level are selected for the two-dimensional image, i is an imaginary symbol, and l=3 is the single highest order in the polynomial model. By using an L-order polynomial phase model, 4 motion phases can be obtained by fitting the motion phases of the measured data through phase unwrapping and phase wrapping operations>The target multi-shot diffusion weighted data to be reconstructed is four shots.
In step 3), the generated simulated diffusion magnetic resonance amplitude map is generatedHigh-order simulation motion phase diagramAnd an acquired multichannel coil sensitivity map +.>Combining according to the following formula to obtain fully sampled network training label data:
wherein,is +.>Or in a non-Cartesian coordinate system>Fourier transform operator of>Is synthesized 8-channel 4-shot fully sampled k-space data as network training tag data.
In step 4), gaussian noise is added to the fully sampled training tag data to obtain noisy k-space data:
wherein,is Gaussian noise with real and imaginary parts meeting the mean value of 0 and the variance of 0.05,is noisy fully sampled k-space data with a signal-to-noise ratio (dB) of X GT Calculated as 10 for the noiseless reference map.
In step 5), setting a sampling track of an acquisition sequence of high-definition diffusion data of a target as a four-time excitation staggered plane echo sequence, undersampling noisy k-space data to obtain simulation 8-channel 4-time excitation diffusion magnetic resonance data of motion artifact interference, wherein the simulation 8-channel 4-time excitation diffusion magnetic resonance data is used as input data of network training:
wherein,the simulation multichannel multi-excitation diffusion magnetic resonance data of motion artifact interference are used as input data of network training. Sampling template->The planar echo sequence (4 shot-rsEPI) of the dimension segment can be read for four excitations, as shown in fig. 2 (b).
In step 6), repeating steps 1) to 5) 100000 times, each time setting different parameters b, A lm μ, σ. Generating 100000 group pairs of network training input data X inp And full sample training tag data X GT Thereby constructing a physical intelligence based high quality paired high definition diffusion imaging training dataset containing 100000 samples. Fig. 3 (b) illustrates a pair of physically intelligent generated network training tags and input band motion artifact data.
In step 7), the generated high quality paired high definition diffusion imaging training dataset of physical intelligence can be used to train a neural network. A convolutional neural network is designed that contains 5 identical reconstruction modules, each comprising 6 convolutional layers, each comprising 32 convolutional kernels of 3 x 3 size and a layer of linear rectifying activation functions (the sixth layer does not contain a linear rectifying activation function). After training, the convolutional neural network can be used for reconstructing actual measurement 4-time excitation high-definition diffusion imaging data.
The physical intelligent data generation specific scheme can rapidly generate paired high-definition diffusion imaging data with multiple b values and multiple resolutions, breaks through the bottleneck that the existing intelligent high-definition diffusion imaging network depends on actual measured data and is lack of high-quality training labels, and has wide application prospect
Example 3
The brain diffusion data generation and denoising network training process based on single excitation planar echo sampling is as follows:
the embodiment of the invention comprises the following steps:
1) Generating a simulation high-definition diffusion magnetic resonance amplitude diagram by using diffusion physical models such as diffusion indexes, tensors, kurtosis and the like;
2) Generating a high-order simulation motion phase diagram of each excitation diffusion image based on the physical motion model;
3) Combining the generated simulation high-definition diffusion magnetic resonance amplitude diagram, the high-order simulation motion phase diagram and the acquired multichannel coil sensitivity map, and transforming the simulation high-definition diffusion magnetic resonance amplitude diagram, the high-order simulation motion phase diagram and the acquired multichannel coil sensitivity map into k space by utilizing Fourier transformation to obtain fully-sampled network training label data;
4) Adding Gaussian noise into the full-sampling training label data to obtain noisy k-space data;
5) Undersampling noisy k-space data according to a sampling track of an acquisition sequence of high-definition diffusion data of a target to obtain simulated multichannel multi-excitation diffusion magnetic resonance data with motion artifact interference, wherein the simulated multichannel multi-excitation diffusion magnetic resonance data is used as input data of network training;
6) Repeating steps 1) to 5) in a large number to generate a large number of paired network training input data and full-sampling training label data;
7) Training a neural network by using the generated diffusion magnetic resonance data set, wherein the network is used for reconstructing or denoising the high-definition diffusion magnetic resonance data after training.
In step 1), the generated simulated high definition diffusion magnetic resonance amplitude map may be generated by a diffusion tensor physical model:
(diffusion tensor model) m (b, g) =exp { -bg T Dg}m 0 (8)
Wherein,for diffusion amplitude image with diffusion b value b in diffusion gradient direction g, +.>For a diffusion image without applied diffusion gradient, T is a transposed symbol, D is a diffusion tensor matrix, and the diffusion gradient is represented by acquiring 1 b value of 0s/mm 2 M of (2) 0 And 64 diffusion direction numbers b of 1000s/mm 2 The actual acquired diffusion image of (2) is calculated by a linear fitting algorithm.
Thus, given a diffusion direction g of [0, 1 ] given D is known]And b has a value b of 1000s/mm 2 A diffusion amplitude image m (g= [0, 1) of the target diffusion direction and b value can be generated],b＝1000)。
In step 2), the high-order simulated phase of the diffusion image may be generated by a physical model:
wherein the diagonal matrixIs the phase obtained by simulation through the model. Integer x, y E [1,128 ]]Coordinates at the level are selected for the two-dimensional image, i being an imaginary symbol. L=3 is the single highest order in the polynomial model. And fitting the motion phase of the measured data through phase unwrapping and phase crimping operation by using a 3-order polynomial phase model to obtain the phase.
In step 3), the generated simulated diffusion magnetic resonance amplitude map is generatedHigh-order simulation motion phase diagramAnd an acquired multichannel coil sensitivity map +.>Combining according to the following formula to obtain fully sampled network training label data:
wherein,is +.>Or in a non-Cartesian coordinate system>Fourier transform operator of>Is synthesized 8-channel fully sampled k-space data as network training tag data.
In step 4), gaussian noise is added to the fully sampled training tag data to obtain noisy k-space data:
wherein,is Gaussian noise with real and imaginary parts meeting the mean value of 0 and the variance of 0.05,is noisy fully sampled k-space data with a signal-to-noise ratio of X GT Calculated for the noiseless reference diagram is 10dB.
The generated physical intelligent data in this embodiment is not undersampled through k-space, thus skipping step 5).
In step 6), repeating steps 1) to 4) 100000 times, each time setting different parameters b, A lm μ, σ. Generating 100000 group pairs of network training input dataAnd full sample training tag data X GT Thereby constructing a physical intelligence based high quality paired high definition diffusion imaging training dataset containing 100000 samples. Fig. 3 (c) shows a pair of physically intelligent generated noise-free high signal-to-noise ratio network training tags and input noisy data.
In step 7), the generated high quality paired high definition diffusion imaging training dataset of physical intelligence can be used to train a neural network. A convolutional neural network is designed that contains 3 identical denoising modules, each denoising module containing 6 convolutional layers, each convolutional layer containing 32 convolutional kernels of 3 x 3 size and a linear rectification activation function layer (the sixth layer contains no linear rectification activation function). After training, the convolutional neural network can be used for denoising actually measured single-shot diffusion imaging data.
The physical intelligent data generation specific scheme can rapidly generate paired high-definition diffusion imaging data with multiple b values and multiple resolutions, breaks through the bottleneck that the existing intelligent high-definition diffusion imaging network depends on actual measurement data and is lack of high-quality training labels, and has wide application prospects.
Reference is made to:
[1]V.Baliyan et al.,“Diffusion weighted imaging:technique and applications,”World Journal of Radiology,8,785,2016.
[2]H.An,X.Ma,Z.Pan,H.Guo,E.Y.P.Lee,“Qualitative and quantitative comparison of image quality between single-shot echo-planar and interleaved multi-shot echo-planar diffusion-weighted imaging in female pelvis,”European Radiology,30,1876-1884,2020.
[3]A.W.Anderson,J.C.Gore,“Analysis and correction of motion artifacts in diffusion weighted imaging,”Magnetic Resonance in Medicine,32,379-387,1994.
[4]Aggarwal.H.K.,M.Mani,M.Jacob,“MoDL-MUSSELS:Model-based deep learning for multi-shot sensitivity-encoded diffusion MRI”,IEEE Transactions on Medical Imaging,39,1268-1277,2019.
[5]C.Qian et al.,“A paired phase and magnitude reconstruction for advanced diffusion-weighted imaging,”IEEE Transactions on Biomedical Engineering,DOI:10.1109/TBME.2023.3288031,2023.
[6]F.Wang et al.,“Multiple b-value model-based residual network(MORN)for accelerated high-resolutiondiffusion-weighted imaging”,IEEE Journal of Biomedical and Health Informatics,26,4575-4586,2022.
[7]Q.Yang,Z.Wang,K.Guo,C.Cai,and X.Qu,“Physics-driven synthetic data learning for biomedical magnetic resonance:The imaging physics based data synthesis paradigm for artificial intelligence,”IEEE Signal Processing Magazine,40,129-140,2023.
[8]C.Qian et al.,“Physics-informed deep diffusion reconstruction:Break the bottleneck of training data in artificial intelligence”,arXiv:2210.11388,2023.
Claims (9)
1. the multi-part physical intelligent high-definition diffusion magnetic resonance data generation method is characterized by comprising the following steps of:
1) Generating a simulation high-definition diffusion magnetic resonance amplitude diagram by using a diffusion index, tensor and kurtosis diffusion physical model;
2) Generating a high-order simulation motion phase diagram of each excitation diffusion image based on the physical motion model;
3) Combining the generated simulation high-definition diffusion magnetic resonance amplitude diagram, the high-order simulation motion phase diagram and the acquired multichannel coil sensitivity map, and transforming the simulation high-definition diffusion magnetic resonance amplitude diagram, the high-order simulation motion phase diagram and the acquired multichannel coil sensitivity map into k space by utilizing Fourier transformation to obtain fully-sampled network training label data;
4) Adding Gaussian noise into the full-sampling training label data to obtain noisy k-space data;
5) Undersampling noisy k-space data according to a sampling track of an acquisition sequence of high-definition diffusion data of a target to obtain simulated multichannel multi-excitation diffusion magnetic resonance data with motion artifact interference, wherein the simulated multichannel multi-excitation diffusion magnetic resonance data is used as input data of network training;
6) Repeating steps 1) to 5) in a large number to generate a large number of paired network training input data and full-sampling training label data;
7) Training a neural network by using the generated diffusion magnetic resonance data set, wherein the network is used for reconstructing or denoising the high-definition diffusion magnetic resonance data after training.
2. The method for generating multi-site physical intelligent high-definition diffusion magnetic resonance data according to claim 1, wherein in step 1), the generated simulated high-definition diffusion magnetic resonance amplitude map is generated by a diffusion index, tensor and kurtosis diffusion physical model:
diffusion single-index model: m (b) =exp { -bdadc } m 0 (1)
Wherein,for diffusion magnitude image with diffusion b value b, < ->For a diffusion image with no diffusion gradient applied, ADC is an apparent diffusion coefficient matrix;
diffusion tensor model: m (b, g) =exp { -bg T Dg}m 0 (2)
Wherein,for diffusion amplitude image with diffusion b value b in diffusion gradient direction g, +.>For a diffusion image to which a diffusion gradient is not applied, g represents a unit vector of a diffusion gradient direction, T is a transpose symbol, and D is a diffusion tensor matrix;
diffusion kurtosis model:
wherein,for diffusion amplitude image with diffusion b value b in diffusion gradient direction g, +.>For a diffusion image to which a diffusion gradient is not applied, g represents a unit vector of a diffusion gradient direction, D is a diffusion tensor matrix, and K is a diffusion kurtosis matrix.
3. The method for generating the multi-part physical intelligent high-definition diffusion magnetic resonance data according to claim 2, wherein constants of ADC, D and K in a diffusion single-index, tensor and kurtosis physical model with tissue diffusion information are calculated by acquiring actual diffusion images with enough diffusion b values and diffusion direction numbers through algorithms such as linear fitting;
given the diffusion direction g and b values b, a diffusion magnitude image m of the target diffusion direction and b values is generated, given the ADC, D, K constants are known.
4. The multi-site physical intelligent high definition diffusion magnetic resonance data generating method as claimed in claim 1, wherein in step 2), the high order simulation motion phases of each excitation diffusion image are generated by a physical motion model:
wherein the diagonal matrixIs the phase obtained by simulation of the model, N and M are the lengths of the frequency code and the phase code of the image, x and y are the coordinates on the selected layer of the two-dimensional image, i is the imaginary symbol, L is the highest order of the single expressions in the polynomial model, M and L-M are the powers of the single expressions x and y in the current first order polynomial, A lm Is the corresponding item x m y l-m Coefficients of (2); fitting the motion phases of the measured data by phase unwrapping and phase crimping operations using an L-th order polynomial phase model to obtain J motion phases +.>J is the excitation times of the multi-excitation diffusion weighted data to be reconstructed; then according to the distribution rule of polynomial coefficients in the L-order polynomial phase model, a large number of polynomial coefficients meeting the distribution rule are obtained, and the groupForming a simulation phase set; the motion conditions of different parts are different, the brain and the prostate part are approximately modeled as rigid motion, and a motion model with L=3 is used for simulating motion phases; the abdominal liver and pancreas regions have more complex elastic motions, and motion models of l=7 or higher order are used to approximate simulated motion phases.
5. The method for generating multi-region physical intelligent high-definition diffusion magnetic resonance data as claimed in claim 1, wherein in step 3), the generated simulated high-definition diffusion magnetic resonance amplitude map is generatedHigh-order simulation motion phase diagram->And an acquired multichannel coil sensitivity map +.>Combining according to the following formula to obtain fully sampled network training label data:
wherein,is +.>Or in a non-Cartesian coordinate system>Fourier transform operator of>Is synthesized H channel J-time excited full-sampling k-space data as network training label data.
6. The method for generating multi-part physical intelligent high-definition diffusion magnetic resonance data according to claim 1, wherein in step 4), gaussian noise is added to the fully sampled training tag data to obtain noisy k-space data as follows:
wherein,is Gaussian noise with real and imaginary parts satisfying mean value mu and variance sigma, < >>Is noisy fully sampled k-space data with a signal-to-noise ratio (dB) of X GT And calculating to obtain the noiseless reference diagram.
7. The method for generating multi-part physical intelligent high-definition diffusion magnetic resonance data according to claim 1, wherein in step 5), the noisy k-space data is undersampled according to a sampling track of an acquisition sequence of target high-definition diffusion data to obtain simulated multi-channel multi-excitation diffusion magnetic resonance data with motion artifact interference, and the simulated multi-channel multi-excitation diffusion magnetic resonance data is used as input data of network training:
wherein,is simulated multichannel multi-excitation diffusion magnetic resonance data of motion artifact interference, is taken as input data of network training, and is sampledForm->Sampling tracks for multiple multi-excitation diffusion sequences in Cartesian coordinate systemSampling template->Selecting a multi-excitation staggered plane echo sequence, and multi-excitation reading-out dimension segmentation plane echo sequence; in the case of a non-Cartesian coordinate system +.>Sampling template->The multi-excitation screw sampling sequence and the multi-excitation screw sampling sequence are selected, and the excitation times are set to J times.
8. The method for generating multi-part physical intelligent high definition diffusion magnetic resonance data as claimed in claim 1, wherein in step 6), repeating steps 1) to 5) K times generates K component pairs of network training input data X inp And full sample training tag data X GT Thereby constructing a high-quality paired high-definition diffusion imaging training data set based on physical intelligence and containing K samples.
9. The multi-site physical intelligent high-definition diffusion magnetic resonance data generation method of claim 1, wherein in step 7), the generated diffusion magnetic resonance data set is used for training a neural network, and after training, the network is used for reconstructing or denoising the high-definition diffusion magnetic resonance data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311004982.7A CN117011409B (en) | 2023-08-10 | 2023-08-10 | Multi-position physical intelligent high-definition diffusion magnetic resonance data generation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311004982.7A CN117011409B (en) | 2023-08-10 | 2023-08-10 | Multi-position physical intelligent high-definition diffusion magnetic resonance data generation method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117011409A true CN117011409A (en) | 2023-11-07 |
| CN117011409B CN117011409B (en) | 2024-05-10 |
Family
ID=88575923
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311004982.7A Active CN117011409B (en) | 2023-08-10 | 2023-08-10 | Multi-position physical intelligent high-definition diffusion magnetic resonance data generation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117011409B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104574298A (en) * | 2014-12-25 | 2015-04-29 | 天津大学 | Multi-b-value DWI (diffusion weighted image) noise reduction method based on mutual information |
| CN108459289A (en) * | 2018-01-30 | 2018-08-28 | 奥泰医疗系统有限责任公司 | A kind of multiple excitation Diffusion weighted MR imaging method based on data consistency |
| WO2021082103A1 (en) * | 2019-10-31 | 2021-05-06 | 清华大学 | Distortionless single-shot echo planar imaging method and apparatus employing deep learning |
| CN113256749A (en) * | 2021-04-20 | 2021-08-13 | 南昌大学 | Rapid magnetic resonance imaging reconstruction algorithm based on high-dimensional correlation prior information |
| CN115471580A (en) * | 2022-08-29 | 2022-12-13 | 厦门大学 | Physical intelligent high-definition magnetic resonance diffusion imaging method |
| US20230236271A1 (en) * | 2022-01-21 | 2023-07-27 | Regents Of The University Of Michigan | Systems and methods for accelerated magnetic resonance imaging (mri) reconstruction and sampling |
-
2023
- 2023-08-10 CN CN202311004982.7A patent/CN117011409B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104574298A (en) * | 2014-12-25 | 2015-04-29 | 天津大学 | Multi-b-value DWI (diffusion weighted image) noise reduction method based on mutual information |
| CN108459289A (en) * | 2018-01-30 | 2018-08-28 | 奥泰医疗系统有限责任公司 | A kind of multiple excitation Diffusion weighted MR imaging method based on data consistency |
| WO2021082103A1 (en) * | 2019-10-31 | 2021-05-06 | 清华大学 | Distortionless single-shot echo planar imaging method and apparatus employing deep learning |
| CN113256749A (en) * | 2021-04-20 | 2021-08-13 | 南昌大学 | Rapid magnetic resonance imaging reconstruction algorithm based on high-dimensional correlation prior information |
| US20230236271A1 (en) * | 2022-01-21 | 2023-07-27 | Regents Of The University Of Michigan | Systems and methods for accelerated magnetic resonance imaging (mri) reconstruction and sampling |
| CN115471580A (en) * | 2022-08-29 | 2022-12-13 | 厦门大学 | Physical intelligent high-definition magnetic resonance diffusion imaging method |
Non-Patent Citations (2)
| Title |
|---|
| CHEN QIAN ET AL.: "Physics-informed Deep Diffusion MRI Reconstruction: Break the Bottleneck of Training Data in Artificial Intelligence", 2023 IEEE 20TH INTERNATIONAL SYMPOSIUM ON BIOMEDICAL IMAGING, 1 September 2023 (2023-09-01) * |
| 丛超: "基于多模态医学影像智能分析的深度学习算法研究与应用", 中国知网, 1 May 2023 (2023-05-01) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117011409B (en) | 2024-05-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Sun et al. | A deep information sharing network for multi-contrast compressed sensing MRI reconstruction | |
| Tezcan et al. | MR image reconstruction using deep density priors | |
| Ghodrati et al. | MR image reconstruction using deep learning: evaluation of network structure and loss functions | |
| Wen et al. | Transform learning for magnetic resonance image reconstruction: From model-based learning to building neural networks | |
| Zeng et al. | A comparative study of CNN-based super-resolution methods in MRI reconstruction and its beyond | |
| Zhou et al. | Parallel imaging and convolutional neural network combined fast MR image reconstruction: Applications in low‐latency accelerated real‐time imaging | |
| Luo et al. | Bayesian MRI reconstruction with joint uncertainty estimation using diffusion models | |
| US10101424B2 (en) | Robust principal component analysis for separation of on and off-resonance in 3D multispectral MRI | |
| Liu et al. | High-performance rapid MR parameter mapping using model-based deep adversarial learning | |
| Pawar et al. | A deep learning framework for transforming image reconstruction into pixel classification | |
| Lai et al. | Joint sparse reconstruction of multi-contrast MRI images with graph based redundant wavelet transform | |
| CN117223028A (en) | System and method for magnetic resonance image reconstruction with denoising | |
| Wang et al. | Denoising auto-encoding priors in undecimated wavelet domain for MR image reconstruction | |
| Li et al. | High quality and fast compressed sensing MRI reconstruction via edge-enhanced dual discriminator generative adversarial network | |
| Oh et al. | A k‐space‐to‐image reconstruction network for MRI using recurrent neural network | |
| Lv et al. | Parallel imaging with a combination of sensitivity encoding and generative adversarial networks | |
| CN114119791A (en) | MRI (magnetic resonance imaging) undersampled image reconstruction method based on cross-domain iterative network | |
| CN114820849A (en) | Magnetic resonance CEST image reconstruction method, device and equipment based on deep learning | |
| Ekanayake et al. | Multi-branch Cascaded Swin Transformers with Attention to k-space Sampling Pattern for Accelerated MRI Reconstruction | |
| Guan et al. | Magnetic resonance imaging reconstruction using a deep energy‐based model | |
| Qian et al. | Physics-informed deep diffusion MRI reconstruction: Break the bottleneck of training data in artificial intelligence | |
| Islam et al. | Compressed sensing regularized calibrationless parallel magnetic resonance imaging via deep learning | |
| CN109188327B (en) | Magnetic resonance image fast reconstruction method based on tensor product complex small compact framework | |
| Zong et al. | Fast reconstruction of highly undersampled MR images using one and two dimensional principal component analysis | |
| CN117011409B (en) | Multi-position physical intelligent high-definition diffusion magnetic resonance data generation method |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |