CN109597012B - Single-scanning space-time coding imaging reconstruction method based on residual error network - Google Patents
Single-scanning space-time coding imaging reconstruction method based on residual error network Download PDFInfo
- Publication number
- CN109597012B CN109597012B CN201811580826.4A CN201811580826A CN109597012B CN 109597012 B CN109597012 B CN 109597012B CN 201811580826 A CN201811580826 A CN 201811580826A CN 109597012 B CN109597012 B CN 109597012B
- Authority
- CN
- China
- Prior art keywords
- space
- imaging
- time coding
- residual error
- magnetic resonance
- 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
- 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
- G01R33/56—Image enhancement or correction, e.g. subtraction or averaging techniques, e.g. improvement of signal-to-noise ratio and resolution
- G01R33/5608—Data processing and visualization specially adapted for MR, e.g. for feature analysis and pattern recognition on the basis of measured MR data, segmentation of measured MR data, edge contour detection on the basis of measured MR data, for enhancing measured MR data in terms of signal-to-noise ratio by means of noise filtering or apodization, for enhancing measured MR data in terms of resolution by means for deblurring, windowing, zero filling, or generation of gray-scaled images, colour-coded images or images displaying vectors instead of pixels
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Radiology & Medical Imaging (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Artificial Intelligence (AREA)
- High Energy & Nuclear Physics (AREA)
- Condensed Matter Physics & Semiconductors (AREA)
- General Physics & Mathematics (AREA)
- Magnetic Resonance Imaging Apparatus (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
A single-scanning space-time coding imaging reconstruction method based on a residual error network relates to a magnetic resonance image reconstruction technology based on a deep learning network. The method for reconstructing the single-scanning space-time coding imaging based on the residual error network is provided, wherein a two-dimensional image is obtained in a single scanning, and then the reconstruction is carried out by using a deep learning method. The excitation pulse is replaced by the linear sweep pulse, so that image distortion caused by an uneven magnetic field and chemical shift is effectively resisted, and the imaging speed, the resolution and the signal-to-noise ratio similar to those of the EPI are obtained. The SPEN imaging is undersampled along the phase encode direction. Although the spatio-temporally encoded imaging signal itself can reflect the contours of the imaged object without reconstruction, the inherent resolution of the contours is typically low. The SPEN image is reconstructed from the signal space with low resolution by utilizing deep learning, the image resolution is greatly improved, proton density distribution is presented, and high signal-to-noise ratio is obtained while the resolution similar to that of the traditional deconvolution reconstruction method is obtained.
Description
Technical Field
The invention relates to a magnetic resonance image reconstruction technology based on a deep learning network, in particular to a single-scanning space-time coding imaging reconstruction technology based on a residual error network.
Background
Magnetic Resonance Imaging (MRI) is an imaging technique that does not destructively analyze structural information of tissue within an object. In the clinic, MRI plays an extremely important role in neuroimaging, cardiovascular imaging, and functional magnetic resonance imaging. Magnetic resonance imaging is performed by applying radio frequency pulses of a certain frequency to the protons while detecting the resulting signals in conjunction with an applied three-dimensional gradient field. Conventional multi-scan magnetic resonance sequences typically take several minutes, even tens of minutes, to obtain a magnetic resonance image, which is unsuitable for clinical applications of MRI. Therefore, the ultra-fast imaging technology can shorten the sampling time, improve the sampling efficiency, improve the image quality to a certain extent and weaken the influence of the motion artifact on the image. With the fast-reading development of ultra-fast imaging technology, planar Echo (EPI) acquisition methods based on Mansfield can acquire complete k-space [1] in a single scan, and are widely applied to perfusion imaging, dynamic imaging and functional imaging. In recent years, a new space-time coding (SPEN) -based imaging technique has been proposed, which can acquire a two-dimensional image with an imaging speed close to that of EPI [2 ]. In the k-space of SPEN, the original imaging dimension of the phase encoding is replaced by the direct position space dimension, i.e. position space information can be obtained only by one-dimensional fourier transform. SPEN imaging can achieve similar sampling times, resolutions, signal-to-noise ratios, etc. as EPI, while improving the resistance to inhomogeneous fields, especially in the case of fully-refocused pulse design, due to the fact that the signal of the SPEN encoding dimension comes directly from the envelope of the position space.
Since the SPEN scanning method introduces a secondary phase distribution related to a spatial position through the sweep pulse, and combines with the stable phase approximation theorem, a magnetic resonance image can be obtained through simple modulus extraction, but compared with the EPI method through fourier transform, the spatial resolution of a modulus value image is seriously reduced. In 2010, Ben-Eliezer et al found that there was redundancy in the signals sampled by the SPEN method, i.e., there was overlap in the phase stability regions of each sample [3 ]. The existing super-resolution reconstruction algorithm by utilizing signal redundancy comprises a conjugate gradient reconstruction algorithm, a partial Fourier reconstruction algorithm [4], a deconvolution reconstruction algorithm and the like. Although the existing method can carry out super-resolution reconstruction, the running time of the algorithm is long, and the denoising effect is not obvious.
Deep learning, an algorithm for efficiently learning a nonlinear mapping between input data and output data through a plurality of hidden layers, has attracted more and more attention in recent years as GPU performance is rapidly improved. Deep learning networks, particularly convolutional neural networks, are widely used in medical image analysis of various problems, including classification, detection, and segmentation. The prior results show that the method based on the convolutional neural network is far superior to the traditional sparse representation method in the aspect of image super-resolution reconstruction [5 ]. As a representative of the deep neural network model, the residual network can solve the problem of gradient explosion/disappearance when the number of network layers is deep.
The SPEN reconstruction is carried out by using the trained deep learning network, the image reconstruction process can be completed in a short time, image noise is effectively removed, and meanwhile, the resolution similar to that of the traditional deconvolution reconstruction method is obtained.
Reference documents:
[1]Stehling M K,Turner R,Mansfield P.Echo-Planar Imaging:MagneticResonance Imaging in a Fraction of a Second[J].Science,1991,254(5028):43-50.
[2]Shrot Y,Frydman L.Spatially encoded NMR and the acquisition of 2Dmagnetic resonance images within a single scan[J].Journal of MagneticResonance,2005,172(2):179-190.
[3]Ben-Eliezer N,Shrot Y,Frydman L,et al.Parametric analysis of thespatial resolution and signal-to-noise ratio in super-resolvedspatiotemporally encoded(SPEN)MRI[J].Magnetic Resonance in Medicine OfficialJournal of the Society of Magnetic Resonance in Medicine,2014,72(2):418-429.
[4]Chen Y,Li J,Qu X,et al.Partial Fourier transform reconstructionfor single-shot MRI with linear frequency-swept excitation[J].MagneticResonance in Medicine Official Journal of the Society of Magnetic Resonancein Medicine,2013,69(5):1326-1336.
[5]Baumgartner C F,Oktay O,Rueckert D.Fully Convolutional Networks inMedical Imaging:Applications to Image Enhancement and Recognition[J].2017.
disclosure of Invention
The invention aims to provide a single-scanning space-time coding imaging reconstruction method based on a residual error network, which obtains a complete two-dimensional image in a single scanning and then obtains higher resolution by using a deep learning method for reconstruction.
The invention comprises the following steps:
1) the code of the space-time coding imaging sequence is written in the operating software of the magnetic resonance imager, and is debugged and compiled.
2) Preparing an experimental sample, fixing the experimental sample on a sample tube of a magnetic resonance imager, and placing the experimental sample in the middle of a cavity of the magnetic resonance imager;
3) and opening magnetic resonance imager operating software on an operating platform of the magnetic resonance imager, firstly, positioning by using a spin echo sequence, finding a proper imaging area, and determining the size of the region of interest and the stratification information. Then carrying out conventional operations of tuning, shimming, frequency correction and power correction;
4) executing the space-time coding imaging sequence in the step 1), setting sampling bandwidth and related parameters, and sampling the region of interest in the step 3) to obtain K space data;
5) normalizing, zero filling and one-dimensional Fourier transform are carried out on the K space data obtained in the step 4) to obtain data of an image domain;
6) generating random samples in data simulation software, carrying out simulated sampling on the random samples in the software to obtain original SPEN K space data, carrying out normalization processing on the K space data, filling zero, and carrying out one-dimensional Fourier transform on a frequency coding dimension to obtain training data for training a deep learning network;
7) building a residual error network model by adopting a TensorFlow deep learning frame and Python, and setting relevant parameters for training; inputting the training data obtained in the step 6) into a network for training until a residual error network converges and reaches stability to obtain a trained network model, and then reconstructing the image domain data obtained in the step 5) by using the trained network model to obtain reliable single-scanning space-time coding imaging based on the residual error network.
In step 1), the structure of the space-time coding imaging sequence is as follows in sequence: linear sweep excitation pulse with a flip angle of 90 degrees, refocusing pulse with a flip angle of 180 degrees, shift gradient and sampling echo train.
In the step 6), the random sample can be randomly generated in batch by using a computer according to the characteristic distribution of the sample to be tested of the experiment, the random sample contains all the characteristics of the experiment sample, and the instability factor is added, so that the robustness of the network model to the non-ideal experiment environment is improved.
In step 7), the trained network can reconstruct an image only requiring tens of milliseconds, which is much shorter than the reconstruction time of the conventional algorithm.
The invention replaces the excitation pulse with the linear sweep pulse, can effectively resist the image distortion caused by the uneven magnetic field and the chemical shift, and simultaneously obtains the imaging speed, the resolution and the signal-to-noise ratio which are similar to the EPI. The SPEN imaging is undersampled along the phase encode direction. Although the spatio-temporally encoded imaging signal itself can reflect the contours of the imaged object without reconstruction, the inherent resolution of the contours is typically low. According to the invention, a SPEN image is reconstructed from a signal space with low resolution by using a deep learning technology, so that the resolution of the image is greatly improved, the proton density distribution is presented more accurately, and a higher signal-to-noise ratio is obtained while the resolution similar to that of the traditional deconvolution reconstruction method is obtained.
Drawings
FIG. 1 is a diagram of a single scan spatio-temporal coding imaging sequence employed by the present invention.
FIG. 2 is a residual network model used to reconstruct spatio-temporal coded images. In fig. 2, the residual network model contains 3 sub-networks: the system comprises an input network, a residual error learning network and a reconstruction network, wherein the input network takes a real part and an imaginary part of a space-time coding image as input; the residual error learning network is the core for image reconstruction; the reconstruction network is used for reconstructing the characteristic maps into a high-resolution two-dimensional image.
Fig. 3 is an image obtained by reconstructing data of a water model, a lemon and a rat brain respectively by using a conventional deconvolution method and a residual error network method.
Detailed Description
The invention is further described below with reference to the accompanying drawings and specific examples.
As shown in fig. 1 and 2, the steps in the implementation of the present invention are as follows:
1) the code of the space-time coding imaging sequence is written in the operating software of the magnetic resonance imager, and is debugged and compiled.
2) Preparing an experimental sample, fixing the experimental sample on a sample tube of a magnetic resonance imager, and placing the experimental sample in the middle of a cavity of the magnetic resonance imager;
3) and opening magnetic resonance imager operating software on an operating platform of the magnetic resonance imager, firstly, positioning by using a spin echo sequence, finding a proper imaging area, and determining the size of the region of interest and the stratification information. Then carrying out conventional operations of tuning, shimming, frequency correction and power correction;
4) executing the space-time coding imaging sequence in the step 1), setting sampling bandwidth and related parameters, and sampling the region of interest in the step 3) to obtain K space data;
5) normalizing, zero filling and one-dimensional Fourier transform are carried out on the K space data obtained in the step 4) to obtain data of an image domain;
6) generating random samples in data simulation software, carrying out simulated sampling on the random samples in the software to obtain original SPEN K space data, carrying out normalization processing on the K space data, filling zero, and carrying out one-dimensional Fourier transform on a frequency coding dimension to obtain training data for training a deep learning network;
7) building a residual error network model by adopting a TensorFlow deep learning frame and Python, and setting relevant parameters for training; inputting the training data obtained in the step 6) into a network for training until a residual error network converges and reaches stability to obtain a trained network model, and then reconstructing the image domain data obtained in the step 5) by using the trained network model to obtain reliable single-scanning space-time coding imaging based on the residual error network.
Specific examples are given below:
experiments are carried out by using a single-scanning space-time coding magnetic resonance reconstruction method based on a residual error network, and imaging experiments are respectively carried out on a water model, a lemon and a rat brain to verify the feasibility of the invention. The experiment was performed under a nuclear magnetic resonance 7T small animal imager. Placing the prepared sample on a sample bed of the instrument, and placing the sample bed in the middle of a coil of a 7T magnetic resonance imager; and opening magnetic resonance imager operating software on an operating platform of the magnetic resonance imager, firstly, positioning by using a spin echo sequence, finding a proper imaging area, and determining the size of the region of interest and the stratification information. Then tuning, shimming, frequency correction and power correction are carried out; and importing and compiling a running space-time coding imaging sequence in the magnetic resonance imager operating software, and setting the sampling bandwidth and other parameters of the pulse sequence after the compiling is passed. The SPEN sequence is then used for sampling.
And after the data sampling is finished, reconstructing the data according to the steps 5) to 7). And inputting the experimental data obtained by sampling into the trained residual error network, and reconstructing to obtain a reliable image. Meanwhile, the SPEN sampling result is subjected to conventional deconvolution reconstruction for comparison, and a reconstructed image is shown in FIG. 3.
The invention applies the space-time coding sequence to the signal sampling process in the magnetic resonance imaging. Then, the sampling signals are input into a trained residual error network after normalization, zero filling and one-dimensional Fourier transform of frequency dimension, and a high-resolution space-time coding image is obtained through reconstruction. Compared with the traditional deconvolution method, the method has faster reconstruction speed and higher signal-to-noise ratio.
Claims (2)
1. A single scanning space-time coding imaging reconstruction method based on a residual error network is characterized by comprising the following steps:
1) compiling a code of a space-time coding imaging sequence in operating software of the magnetic resonance imager, debugging and compiling;
2) preparing an experimental sample, fixing the experimental sample on a sample tube of a magnetic resonance imager, and placing the experimental sample in the middle of a cavity of the magnetic resonance imager;
3) opening magnetic resonance imaging instrument operation software on an operation table of a magnetic resonance imaging instrument, firstly, positioning by using a spin echo sequence, finding a proper imaging area, and determining the size of the information of the layer selection and the area of interest; then carrying out conventional operations of tuning, shimming, frequency correction and power correction;
4) executing the space-time coding imaging sequence in the step 1), setting sampling bandwidth and related parameters, and sampling the region of interest in the step 3) to obtain K space data; the structure of the space-time coding imaging sequence is as follows in sequence: linear sweep frequency excitation pulse with a flip angle of 90 degrees, refocusing pulse with a flip angle of 180 degrees, shift gradient and sampling echo chain;
5) normalizing, zero filling and one-dimensional Fourier transform are carried out on the K space data obtained in the step 4) to obtain data of an image domain;
6) generating random samples in data simulation software, carrying out simulated sampling on the random samples in the software to obtain original SPEN K space data, carrying out normalization processing on the K space data, filling zero, and carrying out one-dimensional Fourier transform on a frequency coding dimension to obtain training data for training a deep learning network;
7) building a residual error network model by adopting a TensorFlow deep learning frame and Python, and setting relevant parameters for training; inputting the training data obtained in the step 6) into a network for training until a residual error network converges and reaches stability to obtain a trained network model, and then reconstructing the image domain data obtained in the step 5) by using the trained network model to obtain reliable single-scanning space-time coding imaging based on the residual error network.
2. The single-scan space-time coding imaging reconstruction method based on the residual error network as claimed in claim 1, wherein in step 6), the random samples are randomly generated in batches by using a computer according to the feature distribution of the sample to be tested of the experiment, the random samples contain all the features of the experimental samples, and the instability factors are added to improve the robustness of the network model to the non-ideal experimental environment.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811580826.4A CN109597012B (en) | 2018-12-24 | 2018-12-24 | Single-scanning space-time coding imaging reconstruction method based on residual error network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811580826.4A CN109597012B (en) | 2018-12-24 | 2018-12-24 | Single-scanning space-time coding imaging reconstruction method based on residual error network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109597012A CN109597012A (en) | 2019-04-09 |
CN109597012B true CN109597012B (en) | 2020-08-04 |
Family
ID=65964262
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811580826.4A Active CN109597012B (en) | 2018-12-24 | 2018-12-24 | Single-scanning space-time coding imaging reconstruction method based on residual error network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109597012B (en) |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110109036B (en) * | 2019-05-24 | 2020-07-03 | 厦门大学 | Two-dimensional space-time coding multi-scan magnetic resonance imaging non-Cartesian sampling and reconstruction method |
CN110895320B (en) * | 2019-10-31 | 2021-12-24 | 清华大学 | Deformation-free single-shot planar echo imaging method and device based on deep learning |
CN113391251B (en) * | 2020-03-12 | 2023-05-26 | 上海联影医疗科技股份有限公司 | Magnetic resonance image reconstruction method, device and equipment |
CN112924913B (en) * | 2021-02-02 | 2022-08-12 | 厦门大学 | Space-time coding magnetic resonance imaging super-resolution reconstruction method and system |
CN113077527B (en) * | 2021-03-16 | 2022-11-18 | 天津大学 | Rapid magnetic resonance image reconstruction method based on undersampling |
CN115494439B (en) * | 2022-11-08 | 2023-04-07 | 中遥天地(北京)信息技术有限公司 | Space-time coding image correction method based on deep learning |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107945146A (en) * | 2017-11-23 | 2018-04-20 | 南京信息工程大学 | A kind of space-time Satellite Images Fusion method based on depth convolutional neural networks |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
DE102005036064B4 (en) * | 2005-08-01 | 2007-07-19 | Siemens Ag | Method for phase-related scheduling of the data flow in switched networks |
CN108010100B (en) * | 2017-12-07 | 2020-02-18 | 厦门大学 | Single-scanning magnetic resonance quantitative T based on residual error network2Imaging reconstruction method |
-
2018
- 2018-12-24 CN CN201811580826.4A patent/CN109597012B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107945146A (en) * | 2017-11-23 | 2018-04-20 | 南京信息工程大学 | A kind of space-time Satellite Images Fusion method based on depth convolutional neural networks |
Non-Patent Citations (1)
Title |
---|
基于改进栈式稀疏去噪自编码器的自适应图像去噪;马红强等;《光学学报》;20181031;第38卷(第10期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN109597012A (en) | 2019-04-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109597012B (en) | Single-scanning space-time coding imaging reconstruction method based on residual error network | |
CN108132274B (en) | Echo-planar imaging is without reference scan pattern distortion antidote under non-uniform magnetic field | |
CN108663644B (en) | Single-scan quantitative magnetic resonance T based on residual error network reconstruction2Imaging method | |
US8379951B2 (en) | Auto calibration parallel imaging reconstruction method from arbitrary k-space sampling | |
CN108459289B (en) | Multi-excitation diffusion weighted magnetic resonance imaging method based on data consistency | |
Ramos‐Llordén et al. | High‐fidelity, accelerated whole‐brain submillimeter in vivo diffusion MRI using gSlider‐spherical ridgelets (gSlider‐SR) | |
CN105785298A (en) | High-precision three-dimensional chemical shift imaging method | |
Fang et al. | Submillimeter MR fingerprinting using deep learning–based tissue quantification | |
CN103885017A (en) | Image distortion correction method based on single sweep quadrature space-time coding magnetic resonance imaging | |
Zhang et al. | Accelerated MRI reconstruction with separable and enhanced low-rank Hankel regularization | |
US7495440B2 (en) | Q-space sampling method and diffusion spectrum imaging method employing the same | |
JP2021010408A (en) | Medical data processing apparatus, medical data processing method, medical data processing program, and magnetic resonance imaging apparatus | |
Varela‐Mattatall et al. | Automatic determination of the regularization weighting for wavelet‐based compressed sensing MRI reconstructions | |
Li et al. | A simultaneous multi‐slice T2 mapping framework based on overlapping‐echo detachment planar imaging and deep learning reconstruction | |
Qian et al. | Physics-informed deep diffusion MRI reconstruction: Break the bottleneck of training data in artificial intelligence | |
CN116725515B (en) | Magnetic resonance rapid imaging method | |
Mani et al. | Multi‐band‐and in‐plane‐accelerated diffusion MRI enabled by model‐based deep learning in q‐space and its extension to learning in the spherical harmonic domain | |
Arefeen et al. | Latent signal models: Learning compact representations of signal evolution for improved time‐resolved, multi‐contrast MRI | |
CN111062994B (en) | Reconstruction method and system of chemical exchange saturation transfer contrast image | |
CN110109036B (en) | Two-dimensional space-time coding multi-scan magnetic resonance imaging non-Cartesian sampling and reconstruction method | |
Tobisch et al. | Comparison of basis functions and q‐space sampling schemes for robust compressed sensing reconstruction accelerating diffusion spectrum imaging | |
CN113920211B (en) | Quick magnetic sensitivity weighted imaging method based on deep learning | |
Shimron et al. | CORE‐PI: Non‐iterative convolution‐based reconstruction for parallel MRI in the wavelet domain | |
Li et al. | Sampling strategies and integrated reconstruction for reducing distortion and boundary slice aliasing in high‐resolution 3D diffusion MRI | |
Le et al. | Deep Learning–based Method for Denoising and Image Enhancement in Low-Field MRI |
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 |