CN113192151B - MRI image reconstruction method based on structural similarity - Google Patents
MRI image reconstruction method based on structural similarity Download PDFInfo
- Publication number
- CN113192151B CN113192151B CN202110379886.5A CN202110379886A CN113192151B CN 113192151 B CN113192151 B CN 113192151B CN 202110379886 A CN202110379886 A CN 202110379886A CN 113192151 B CN113192151 B CN 113192151B
- Authority
- CN
- China
- Prior art keywords
- contrast
- mri image
- reconstruction
- image
- sampling
- 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.)
- Active
Links
Images
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
-
- 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
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
Abstract
The invention discloses an MRI image reconstruction method based on structural similarity, and belongs to the technical field of medical image optimization. The method comprises the following steps: carrying out analog undersampling pretreatment on the fully sampled MRI image; establishing a single contrast ratio reconstruction model of the contrast ratio of T1 and T2; constructing a similarity constraint loss function; performing fusion training on a multi-contrast fusion reconstruction model; and inputting the undersampled MRI images with a plurality of contrasts at the same part into a multi-contrast fusion reconstruction model. The method utilizes a deep learning technology with strong fitting capability to establish single-contrast reconstruction models for the MRI images with different contrasts, and uses structural similarity constraint to perform fusion training on the single-contrast reconstruction models with different contrasts, so that a higher-quality reconstruction effect is achieved compared with a single-contrast reconstruction neural network model.
Description
Technical Field
The invention relates to the technical field of medical image optimization, in particular to an MRI image reconstruction method based on structural similarity.
Background
The magnetic resonance imaging technology is an important medical imaging technology in the modern medical field and is widely applied to the work of generating accurate living anatomy images and the like. In a common MRI scan, images with different contrasts, such as T1 contrast and T2 contrast, can be obtained, and in actual diagnosis, doctors need to perform disease diagnosis by combining complete MRI images with multiple contrasts: the T1 contrast is favorable for observing an anatomical structure, and the T2 contrast is favorable for observing the focus of a patient; therefore, the patient needs to be scanned for a plurality of times for a long time to obtain enough data for the diagnosis and analysis of a doctor, but the long-time scanning not only affects the health of the patient, but also may introduce motion artifacts, which is not favorable for the accuracy of image data; while decreasing the scan time will decrease the amount of data acquired, which can lead to a decrease in MRI image quality.
In order to solve this conflict, researchers in the related field have conducted many researches and researches on medical image reconstruction methods using modern computing science.
In the existing research of reconstructing MRI images based on the deep learning technique, researchers propose a method of reconstructing undersampled MRI images by simply using a convolutional neural network [ s.wang et al.a. Acquiring magnetic resonance imaging via discrete reconstruction, 2016], and also improve the model design of the convolutional neural network, and propose a cascaded convolutional neural network for reconstructing MRI images [ j.schlemper, et al.a. deep learning network of a spatial neural network for MR image reconstruction,2017]. However, the existing reconstruction method based on the deep learning MRI image generally only obtains a better reconstruction effect in the reconstruction of the MRI image with single contrast, cannot perform fusion training on MRI image reconstruction networks with different contrasts, and cannot improve the speed and quality of image reconstruction by utilizing the characteristic of structural similarity of the MRI images with different contrasts of the same human body part.
Disclosure of Invention
In order to overcome the defects that the prior art cannot utilize the structural similarity of MRI images with different contrasts of the same human body part and the image reconstruction quality is poor, the invention provides an MRI image reconstruction method based on the structural similarity, and the technical scheme is as follows:
an MRI image reconstruction method based on structural similarity comprises the following steps:
s1, preprocessing a plurality of contrast full-sampling MRI images to obtain a simulated undersampled MRI image;
s2, establishing a plurality of single-contrast reconstruction models, wherein the loss function of each single-contrast reconstruction model is a single-contrast loss function L pri ；
S3, constructing similarity constraint based on the structural similarity of a plurality of single-contrast images at the same part; the method specifically comprises the following steps: designing an image gradient information extraction operator, acquiring edge information of an image by using the image gradient information extraction operator, taking the similarity of MRI image gradient information with different contrasts as constraint conditions, and constructing a similarity constraint loss function L cst ；
S4, utilizing the single-contrast reconstruction model to perform fusion training on the multi-contrast fusion reconstruction modelMolding; the method specifically comprises the following steps: for single contrast loss function L pri And similarity constraint loss function L cst Weighted summation is carried out to obtain the total loss function L of the multi-contrast fusion reconstruction model total (ii) a Carrying out model optimization through a gradient descent method to obtain a multi-contrast fusion reconstruction model;
and S5, inputting the under-sampled MRI images with a plurality of contrasts at the same part into a multi-contrast fusion reconstruction model to obtain the MRI image reconstruction result of each contrast.
Analyzing and verifying gradient information of MRI images with different contrasts of the same human body part to obtain that the corresponding images of the same human body part have structural similarity; according to the characteristic, the technical scheme provides an MRI image reconstruction method based on structural similarity, a single-contrast reconstruction model is established for MRI images with different contrasts by utilizing a deep learning technology with strong fitting capacity, fusion training is carried out on the single-contrast reconstruction models with different contrasts by using structural similarity constraint, a neural network model which can be combined with the structural similarity of the images and can reconstruct two under-sampled MRI images with different contrasts is trained under the drive of a large amount of fully-sampled MRI image data, and compared with the single-contrast reconstruction neural network model, the reconstruction effect with higher quality is achieved.
Further, the preprocessing procedure in step S1 is: firstly, two-dimensional Fourier transform is carried out on original single-channel two-dimensional image data of a fully-sampled MRI image to obtain frequency domain data under K space, random down-sampling is carried out on the frequency domain data under the K space, zero filling is carried out on lost data, and finally two-dimensional inverse Fourier transform is carried out on the data subjected to the zero filling to obtain an analog under-sampling MRI image.
In the above technical solution, the sampling rate is 20%.
Further, in step S2, the plurality of single-contrast reconstruction models include a T1-contrast reconstruction model and a T2-contrast reconstruction model, and the T1-contrast reconstruction model and the T2-contrast reconstruction model are used for constructing the multi-contrast fusion reconstruction model in step S4.
Further, the single contrast reconstruction model function is:
y i ＝Single_recon(x i )
wherein x is i ∈R N Is an undersampled MRI image comprising a T1 contrast undersampled MRI image or a T2 contrast undersampled MRI image, the undersampled MRI image being an input to the model; y is i ∈R N Reconstructing an MRI image for the finally obtained single contrast; single _ recon (-) is a deep convolutional neural network model that maps an undersampled MRI image to a Single contrast reconstructed MRI image.
Further, the single-contrast reconstruction model is constructed by utilizing a convolutional neural network to perform feature extraction on the simulated under-sampled MRI image, and sequentially performing three times of down-sampling operation and three times of up-sampling operation on the simulated under-sampled MRI image so as to maintain the original size of the image; the down-sampling mode is maximum pooling operation, and the up-sampling mode is deconvolution operation; and connecting the output features of the down-sampling maximum pooling layer to the initial input features of the up-sampling deconvolution operation layer, and using the output features and the initial input features as input of the deconvolution layer.
Further, a single contrast loss function L of the single contrast reconstruction model pri Is defined as:
wherein the content of the first and second substances,
is a single contrast undersampled MRI image;
is a single contrast fully sampled MRI image; f u (. The) is a Fourier transform function, and a fully sampled MRI image or a single contrast reconstructed MRI image is converted into a representation form under K space; n is the total number of samples; l is a radical of an alcohol pri1 Is a spatial domain fidelity term; l is pri2 Is a Fourier domain fidelity term, θ 1 And theta 2 Are respectively L pri1 And L pri2 The weight coefficient of (2).
Further, the similarity constraint loss function L of step S3 cst The method comprises the following steps: strong gradient similarity constraint L S Weak gradient similarity constraint L W And guided filtering type similarity constraint L GF The calculation method comprises the following steps:
wherein, the first and the second end of the pipe are connected with each other,
the representative image gradient information extraction operator comprises horizontal gradient information and vertical gradient information; n is the total number of samples; lambda is weakly bound L W The hyper-parameter of (c); w k To guide the filtering sliding window linear transform coefficients, they will vary as the input map, guide map and sliding window positions vary.
Further, the total loss function L of step S4 total The calculation formula of (2) is as follows:
L total ＝ρ1·L pri +ρ2·L S +ρ3·L w +ρ4·L GF
wherein ρ 1, ρ 2, ρ 3, ρ 4 are single contrast loss functions L, respectively pri Strong gradient similarity constraint L S Weak gradient similarity constraint L W And guided filtering type similarity constraint L GF The weight coefficient of (a); l is pri Is a loss function that includes a T1 contrast undersampled MRI image and a T2 contrast undersampled MRI image.
Further, a single contrast loss function L for the T1 contrast reconstruction model pri The under-sampled MRI image and the full-sampled MRI image are both T1 contrast images; single contrast loss function L for T2 contrast reconstruction model pri And the undersampled MRI image and the fully sampled MRI image are both T2 contrast images.
Compared with the prior art, the technical scheme of the invention has the beneficial effects that:
analyzing and verifying gradient information of MRI images with different contrasts of the same human body part to obtain that the corresponding images of the same human body part have structural similarity; according to the characteristic, the invention provides an MRI image reconstruction method based on structural similarity, a single-contrast reconstruction model is established for MRI images with different contrasts by utilizing a deep learning technology with strong fitting capacity, the structural similarity constraint is used for carrying out fusion training on the single-contrast reconstruction models with different contrasts, a neural network model which can be combined with the structural similarity of the images and simultaneously rebuild two undersampled MRI images with different contrasts is trained under the drive of a large amount of fully sampled MRI image data, and compared with the single-contrast reconstruction neural network model, the reconstruction effect with higher quality is achieved.
Drawings
Fig. 1 is a flowchart of an MRI image reconstruction method based on structural similarity.
FIG. 2 is a schematic diagram of a single contrast reconstruction model framework.
FIG. 3 is a schematic diagram of a multi-contrast fusion reconstruction model framework.
Fig. 4 is a flow chart of the analog undersampling preprocessing.
Detailed Description
The drawings are for illustrative purposes only and are not to be construed as limiting the patent;
the technical solution of the present invention is further described with reference to the drawings and the embodiments.
Examples
The invention provides an MRI image reconstruction method based on structural similarity, which has the implementation steps as shown in figure 1 and comprises the following detailed steps:
s1, preprocessing a plurality of contrast full-sampling MRI images to obtain a simulated under-sampling MRI image;
the originally acquired image data is fully sampled MRI image data, the originally acquired image data needs to be preprocessed to obtain under-sampled MRI image data as training data of a model, the fully sampled MRI images with multiple contrasts include T1 contrast fully sampled MRI images and T2 contrast fully sampled MRI images, and the preprocessing process is as shown in fig. 4 and includes: firstly, two-dimensional Fourier transform is carried out on original single-channel two-dimensional image data of a fully-sampled MRI image to obtain frequency domain data under K space, random down-sampling is carried out on the frequency domain data under the K space with the sampling rate of 20%, missing data is filled with zeros, and finally two-dimensional inverse Fourier transform is carried out on the data subjected to the zero-padding to obtain a simulated under-sampled MRI image.
In other embodiments of the present invention, the plurality of contrast MRI images further includes a pd contrast fully sampled MRI image and a diffusion contrast fully sampled MRI image.
S2, establishing a plurality of single-contrast reconstruction models;
the multiple single-contrast reconstruction models in this embodiment include a T1-contrast reconstruction model and a T2-contrast reconstruction model.
The single contrast reconstruction model function is:
y i ＝Single_recon(x i )
wherein x is i ∈R N Is an undersampled MRI image comprising a T1 contrast undersampled MRI image or a T2 contrast undersampled MRI image, the undersampled MRI image being an input to the model; y is i ∈R N Reconstructing an MRI image for the finally obtained single contrast; single _ recon (-) is a deep convolutional neural network model that maps an undersampled MRI image to a Single contrast reconstructed MRI image.
The construction process of the single-contrast reconstruction model comprises the steps of utilizing a convolutional neural network to carry out feature extraction on a simulated undersampled MRI image, and carrying out three times of downsampling operation and three times of upsampling operation on the simulated undersampled MRI image in sequence so as to maintain the original size of the image; the down sampling mode is maximum pooling operation, and the up sampling mode is deconvolution operation; in order to improve the memory capacity of the neural network to the initial features of the image, the output features of the down-sampling maximum pooling layer are connected to the initial input features of the up-sampling deconvolution operation layer and are used as the input of the deconvolution layer.
The single contrast reconstruction model framework is shown in FIG. 2, and the single contrast loss function L of the single contrast reconstruction model under supervised conditions including fully sampled MRI images pri Is defined as:
wherein the content of the first and second substances,
is a single contrast undersampled MRI image;
is a single contrast fully sampled MRI image; f u (. The) is a Fourier transform function, and a fully sampled MRI image or a single contrast reconstructed MRI image is converted into a representation form under K space; n is the total number of samples; l is pri1 Is a spatial domain fidelity term; l is pri2 Is a Fourier domain fidelity term, θ 1 And theta 2 Are each L pri1 And L pri2 The weight coefficient of (2).
Pre _ proc (-) in fig. 2 is a post-processing operation to ensure that the single contrast reconstructed MRI image output by the single contrast reconstruction model is consistent with the K-space value of the original under-sampled MRI image,
MRI images are reconstructed for single contrast of Tj contrast.
In other embodiments of the present invention, the single contrast reconstruction model further comprises a pd contrast reconstruction model and a diffusion contrast reconstruction model.
S3, constructing similarity constraint based on the structural similarity of a plurality of single-contrast images at the same part; the method specifically comprises the following steps: design imageGradient information extraction operator, which is used for obtaining the edge information of the image and constructing a similarity constraint loss function L by taking the similarity of the gradient information of the MRI images with different contrasts as constraint conditions cst ；
The similarity constraint loss function L cst The method comprises the following steps: strong gradient similarity constraint L S Weak gradient similarity constraint L W And guided filtering type similarity constraint L GF The calculation method comprises the following steps:
wherein the content of the first and second substances,
the representative image gradient information extraction operator comprises horizontal gradient information and vertical gradient information; n is the total number of samples; λ is weakly bound L W The hyper-parameter of (c); w is a group of k To guide the linear transformation coefficient of the filter sliding window, the position of the guide graph and the sliding window is changed along with the change of the input graph, the guide graph and the sliding window; strong constraint L S The pixel values corresponding to the gradient information are constrained, so that the method has stronger constraint and ensures local similarity; while weakly constraining L W The gradient information of the two contrast images shows consistent variation trend in the whole situation; designing similarity constraint L using guided filtering principle GF Explanation will be given by taking one case as an example: complete image of T2 contrast undersampled image after deep convolution neural network Single _ recon (-) reconstruction
AsGuiding the gradient information of the filter input map, T1 contrast image
As a guide map, the output of the guide filter and the gradient information of the T2 contrast image are compared
As part of the total loss function.
S4, fusing and training a multi-contrast fusion reconstruction model by using the single-contrast reconstruction model; the method specifically comprises the following steps: the T1 contrast ratio reconstruction model and the T2 contrast ratio reconstruction model are subjected to similarity constraint loss function L cst Constructing to obtain a multi-contrast fusion reconstruction model, and performing a single-contrast loss function L pri And similarity constraint loss function L cst Weighted summation is carried out to obtain the total loss function L of the multi-contrast fusion reconstruction model total (ii) a Model optimization by gradient descent method for finding total loss function L total Outputting to obtain a multi-contrast fusion reconstruction model;
the multi-contrast fusion reconstruction model framework is shown in FIG. 3, and the total loss function L total The calculation formula of (2) is as follows:
L total ＝ρ1·L pri +ρ2·L S +ρ3·L w +ρ4·L GF
wherein ρ 1, ρ 2, ρ 3, ρ 4 are single contrast loss functions L, respectively pri Strong gradient similarity constraint L S Weak gradient similarity constraint L W And guided filtering type similarity constraint L GF The weight coefficient of (a); l is pri And (3) reconstructing single contrast loss functions of models for the two single contrasts of T1 and T2. The total loss function L total Is calculated by the following formula pri Including the loss function of the T1 contrast undersampled MRI image and the T2 contrast undersampled MRI image.
And S5, inputting the under-sampled MRI images with a plurality of contrasts at the same part into a multi-contrast fusion reconstruction model to obtain fusion reconstruction MRI images with each contrast.
The method comprises two parts of model training and image reconstruction, wherein in the subsequent image reconstruction task, if the multi-contrast fusion reconstruction model is trained, the MRI images with multiple contrasts at the same part are directly input into the image reconstruction model to obtain an image reconstruction result. In the embodiment, the image reconstruction model is used for reconstructing an image by combining structural similarity of a T1 contrast MRI image and a T2 contrast MRI image, and in other embodiments of the present invention, MRI images which can be used for combining reconstruction further include a pd contrast MRI image and a diffusion contrast MRI image.
It should be understood that the above-described embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.
Claims (9)
1. An MRI image reconstruction method based on structural similarity is characterized by comprising the following steps:
s1, preprocessing a plurality of contrast full-sampling MRI images to obtain a simulated under-sampling MRI image;
s2, establishing a plurality of single-contrast reconstruction models, wherein the loss function of each single-contrast reconstruction model is a single-contrast loss function L pri ；
S3, constructing similarity constraint based on the structural similarity of a plurality of single-contrast images at the same part; the method specifically comprises the following steps: designing an image gradient information extraction operator, acquiring edge information of an image by using the image gradient information extraction operator, taking the similarity of MRI image gradient information with different contrasts as a constraint condition, and constructing a strong gradient similarity constraint L S Weak gradient similarity constraint L W And guided filtering type similarity constraint L GF Commonly composed similarity constrained loss function L cst ；
S4, fusing and training a multi-contrast fusion reconstruction model by using the single-contrast reconstruction model; the method specifically comprises the following steps: for single contrast loss function L pri And similarity constraint loss function L cst Weighted summation is carried out to obtain the total loss function L of the multi-contrast fusion reconstruction model total (ii) a Carrying out model optimization through a gradient descent method to obtain a multi-contrast fusion reconstruction model;
and S5, inputting the under-sampled MRI images with a plurality of contrasts at the same part into a multi-contrast fusion reconstruction model to obtain the MRI image reconstruction result of each contrast.
2. The method for reconstructing an MRI image based on structural similarity according to claim 1, wherein the preprocessing procedure in step S1 is: firstly, two-dimensional Fourier transform is carried out on original single-channel two-dimensional image data of a fully-sampled MRI image to obtain frequency domain data under K space, random down-sampling is carried out on the frequency domain data under the K space, zero filling is carried out on lost data, and finally two-dimensional inverse Fourier transform is carried out on the data subjected to the zero filling to obtain an analog under-sampling MRI image.
3. The MRI image reconstruction method based on the structural similarity as claimed in claim 1, wherein the plurality of single contrast reconstruction models of step S2 includes a T1 contrast reconstruction model and a T2 contrast reconstruction model, and the T1 contrast reconstruction model and the T2 contrast reconstruction model are used for constructing the multi-contrast fusion reconstruction model of step S4.
4. A method for reconstructing MRI image based on structural similarity according to claim 3, wherein said single contrast reconstruction model function in step S2 is:
y i ＝Single_recon(x i )
wherein x is i ∈R N Is an undersampled MRI image comprising a T1 contrast undersampled MRI image or a T2 contrast undersampled MRI image, the undersampled MRI image being a modelThe input of (1); y is i ∈R N Reconstructing an MRI image for the finally obtained single contrast; single _ recon (-) is a deep convolutional neural network model that maps an undersampled MRI image to a Single contrast reconstructed MRI image.
5. The MRI image reconstruction method based on the structural similarity according to claim 4, characterized in that the single contrast ratio reconstruction model in step S2 is constructed by performing feature extraction on the simulated under-sampled MRI image by using a convolutional neural network, and performing three down-sampling operations and three up-sampling operations on the simulated under-sampled MRI image in sequence to maintain the original size of the image; the down-sampling mode is maximum pooling operation, and the up-sampling mode is deconvolution operation; and connecting the output features of the down-sampling maximum pooling layer to the initial input features of the up-sampling deconvolution operation layer, and using the output features and the initial input features as input of the deconvolution layer.
6. The structural similarity-based MRI image reconstruction method according to claim 5, wherein the step S2 is a single contrast loss function L of the single contrast reconstruction model pri Is defined as follows:
wherein, the first and the second end of the pipe are connected with each other,
the method comprises the steps of obtaining an under-sampled MRI image with single contrast, wherein the under-sampled MRI image comprises a T1 contrast under-sampled MRI image and a T2 contrast under-sampled MRI image;
the MRI image is a single-contrast full-sampling MRI image, and the full-sampling MRI image comprises a T1 contrast full-sampling MRI image and a T2 contrast full-sampling MRI image; f u (. To) a fully sampled MRI image or a single contrast reconstructed MRI image as a down-sampled Fourier transform functionConverting the image into a representation form under a K space and performing down-sampling; n is the total number of samples; l is a radical of an alcohol pri1 Is a spatial domain fidelity term; l is pri2 Is a Fourier domain fidelity term, θ 1 And theta 2 Are respectively L pri1 And L pri2 The weight coefficient of (2).
7. The structural similarity-based MRI image reconstruction method according to claim 6, wherein the similarity constraint loss function L of step S3 cst The method comprises the following steps: strong gradient similarity constraint L S Weak gradient similarity constraint L W And guided filtering type similarity constraint L GF The calculation method comprises the following steps:
wherein, the first and the second end of the pipe are connected with each other,
and
respectively representing a T1 contrast ratio MRI image and a T2 contrast ratio MRI image which are obtained through the primary reconstruction of a corresponding single contrast ratio reconstruction model;
the representative image gradient information extraction operator comprises horizontal gradient information and vertical gradient information; n is the total number of samples; lambda is weakly bound L W The hyper-parameter of (c); w k Sliding window for guided filteringThe linear transformation coefficients will vary as the input map, guide map and sliding window positions vary.
8. A method for reconstructing MRI image based on structural similarity according to claim 1, characterized in that said overall loss function L of step S4 total The calculation formula of (2) is as follows:
L total ＝ρ1·L pri +ρ2·L S +ρ3·L w +ρ4·L GF
wherein ρ 1, ρ 2, ρ 3, ρ 4 are single contrast loss functions L, respectively pri Strong gradient similarity constraint L S Weak gradient similarity constraint L W Similarity constraint L of the guided filtering type GF The weight coefficient of (a); l is pri Is a loss function that includes a T1 contrast undersampled MRI image and a T2 contrast undersampled MRI image.
9. A method for reconstructing MRI images based on structural similarity according to claim 6, characterized in that the single contrast loss function L for the T1 contrast reconstruction model pri The under-sampling MRI image and the full-sampling MRI image are both T1 contrast images; single contrast loss function L for T2 contrast reconstruction model pri And the undersampled MRI image and the fully sampled MRI image are both T2 contrast images.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110379886.5A CN113192151B (en) | 2021-04-08 | 2021-04-08 | MRI image reconstruction method based on structural similarity |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110379886.5A CN113192151B (en) | 2021-04-08 | 2021-04-08 | MRI image reconstruction method based on structural similarity |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113192151A CN113192151A (en) | 2021-07-30 |
| CN113192151B true CN113192151B (en) | 2022-12-27 |
Family
ID=76975158
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110379886.5A Active CN113192151B (en) | 2021-04-08 | 2021-04-08 | MRI image reconstruction method based on structural similarity |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113192151B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023115339A1 (en) * | 2021-12-21 | 2023-06-29 | 深圳先进技术研究院 | Physics-based data augmentation contrastive learning representation imaging method |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109410289A (en) * | 2018-11-09 | 2019-03-01 | 中国科学院武汉物理与数学研究所 | A kind of high lack sampling hyperpolarized gas lung MRI method for reconstructing of deep learning |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108090871B (en) * | 2017-12-15 | 2020-05-08 | 厦门大学 | Multi-contrast magnetic resonance image reconstruction method based on convolutional neural network |
| US11175365B2 (en) * | 2018-10-02 | 2021-11-16 | General Electric Company | System and method for sparse image reconstruction utilizing null data consistency |
| CN111696168B (en) * | 2020-06-13 | 2022-08-23 | 中北大学 | High-speed MRI reconstruction method based on residual self-attention image enhancement |
| CN111870245B (en) * | 2020-07-02 | 2022-02-11 | 西安交通大学 | Cross-contrast-guided ultra-fast nuclear magnetic resonance imaging deep learning method |
| CN112489155B (en) * | 2020-12-07 | 2024-01-26 | 中国科学院深圳先进技术研究院 | Image reconstruction method and device, electronic equipment and machine-readable storage medium |
-
2021
- 2021-04-08 CN CN202110379886.5A patent/CN113192151B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109410289A (en) * | 2018-11-09 | 2019-03-01 | 中国科学院武汉物理与数学研究所 | A kind of high lack sampling hyperpolarized gas lung MRI method for reconstructing of deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113192151A (en) | 2021-07-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108460726B (en) | Magnetic resonance image super-resolution reconstruction method based on enhanced recursive residual network | |
| CN111081354B (en) | System and method for denoising medical images through deep learning network | |
| CN113077527B (en) | Rapid magnetic resonance image reconstruction method based on undersampling | |
| CN110461228A (en) | The quality of medical image is improved using more contrasts and deep learning | |
| CN108629816B (en) | Method for reconstructing thin-layer magnetic resonance image based on deep learning | |
| US10852376B2 (en) | Magnetic resonance imaging method and device | |
| CN106485764B (en) | The quick exact reconstruction methods of MRI image | |
| CN111951344B (en) | Magnetic resonance image reconstruction method based on cascade parallel convolution network | |
| CN113298710B (en) | Optical coherence tomography super-resolution imaging method based on external attention mechanism | |
| Du et al. | Accelerated super-resolution MR image reconstruction via a 3D densely connected deep convolutional neural network | |
| CN111487573B (en) | Enhanced residual error cascade network model for magnetic resonance undersampling imaging | |
| CN106618571A (en) | Nuclear magnetic resonance imaging method and system | |
| CN113192151B (en) | MRI image reconstruction method based on structural similarity | |
| CN111784792A (en) | Rapid magnetic resonance reconstruction system based on double-domain convolution neural network and training method and application thereof | |
| Belov et al. | Towards ultrafast MRI via extreme k-space undersampling and superresolution | |
| CN114549318A (en) | Ultra-high resolution fluorescence microscopic imaging method based on sub-voxel convolution neural network | |
| Dehner et al. | A deep neural network for real-time optoacoustic image reconstruction with adjustable speed of sound | |
| CN117576240A (en) | Magnetic resonance image reconstruction method based on double-domain transducer | |
| CN109741439A (en) | A kind of three-dimensional rebuilding method of two dimension MRI fetus image | |
| CN112085810A (en) | Brain tissue free water imaging reconstruction method and system, storage medium and terminal | |
| CN106228583A (en) | The device that a kind of abdomen images is rebuild | |
| CN113838105B (en) | Diffusion microcirculation model driving parameter estimation method, device and medium based on deep learning | |
| CN112489150B (en) | Multi-scale sequential training method of deep neural network for rapid MRI | |
| CN114140404A (en) | Lung multi-core MRI (magnetic resonance imaging) double-domain super-resolution reconstruction method based on artificial intelligence | |
| Hammernik et al. | Deep Learning for Parallel MRI Reconstruction: Overview, Challenges, and Opportunities |
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 |