WO2022213666A1 - 结合k空间和图像空间重建的成像方法和装置 - Google Patents

结合k空间和图像空间重建的成像方法和装置 Download PDF

Info

Publication number
WO2022213666A1
WO2022213666A1 PCT/CN2021/140280 CN2021140280W WO2022213666A1 WO 2022213666 A1 WO2022213666 A1 WO 2022213666A1 CN 2021140280 W CN2021140280 W CN 2021140280W WO 2022213666 A1 WO2022213666 A1 WO 2022213666A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
space
frame
image frame
undersampled
Prior art date
Application number
PCT/CN2021/140280
Other languages
English (en)
French (fr)
Inventor
张祎
祖涛
吴丹
Original Assignee
浙江大学
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 浙江大学 filed Critical 浙江大学
Priority to JP2023558119A priority Critical patent/JP2024512529A/ja
Publication of WO2022213666A1 publication Critical patent/WO2022213666A1/zh
Priority to US18/482,883 priority patent/US20240036141A1/en

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
    • G01R33/561—Image enhancement or correction, e.g. subtraction or averaging techniques, e.g. improvement of signal-to-noise ratio and resolution by reduction of the scanning time, i.e. fast acquiring systems, e.g. using echo-planar pulse sequences
    • G01R33/5611—Parallel magnetic resonance imaging, e.g. sensitivity encoding [SENSE], simultaneous acquisition of spatial harmonics [SMASH], unaliasing by Fourier encoding of the overlaps using the temporal dimension [UNFOLD], k-t-broad-use linear acquisition speed-up technique [k-t-BLAST], k-t-SENSE
    • 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
    • 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
    • G01R33/5605—Image enhancement or correction, e.g. subtraction or averaging techniques, e.g. improvement of signal-to-noise ratio and resolution by transferring coherence or polarization from a spin species to another, e.g. creating magnetization transfer contrast [MTC], polarization transfer using nuclear Overhauser enhancement [NOE]
    • 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

Definitions

  • the invention belongs to the field of magnetic resonance imaging, and can realize fast acquisition for the measurement of magnetic resonance parameters requiring multiple frames of image data.
  • Magnetic resonance imaging can be used to measure tissues including nuclear spin density, longitudinal relaxation time (T1), transverse relaxation time (T2), apparent diffusion coefficient (ADC), cerebral blood flow rate, magnetization transfer rate, etc. parameter.
  • Signal sources for these parameters can be the nucleus of endogenous compounds present in the organism (eg 1 H in water), or exogenous substances or tracers. Direct quantification of tissue MRI parameters can provide more valuable clinical information on various pathologies in neurological, musculoskeletal, liver and myocardial imaging.
  • Imaging techniques to measure these parameters include relaxation time imaging, diffusion imaging, perfusion imaging, functional magnetic resonance imaging (fMRI), and chemical exchange saturation transfer (CEST) imaging, most of which require the acquisition of multiple pulse train parameter modulations on the same imaging subject. frame image dataset.
  • modulated pulse sequence parameters include echo time (TE), flip angle or gradient strength, etc.
  • TE echo time
  • Target parameter estimation can be achieved by fitting the signal evolution in a multi-frame image dataset with a mathematical model. Due to the need for multiple acquisitions, the scanning time is very long, which limits its clinical application; on the other hand, the longer scanning time easily introduces additional motion, which makes the technology itself more sensitive to motion. Therefore, reducing the data acquisition time and speeding up the imaging process is not only convenient for the patient, but also helps to improve the image quality.
  • k-w ROSA can estimate asymmetric Z-spectra of interest directly from complete or incomplete measurements by incorporating subspace-based Z-spectrum signal decomposition into the measurement model, but its speed-up capability is limited.
  • the CS-CEST method based on compressed sensing uses the sparsity of the image in the transform domain to speed up, but has not yet achieved complete random undersampling in practical applications.
  • GRAPPA self-calibrated k-space method
  • SENSE image space method based on dominant coil sensitivity
  • A represents the source data matrix generated by the automatic calibration signal (ACS)
  • y represents the target data vector
  • w represents the weights to be fitted.
  • Previous studies have shown that GRAPPA can produce accurate images with high robustness when the acceleration factor is low. For parameter measurement that requires multiple frames of data, since many frames of images need to be collected, the traditional GRAPPA method needs to collect ACS for each image frame, which greatly reduces the speed-up effect; if only one ACS is collected, the calculated Weights are applied to all undersampled frames, again introducing unknown artifacts.
  • GRAPPA can only collect data from far apart k-spaces to fit missing data, which increases the error.
  • the principle of GRAPPA limits its speed-up capability. In practice, two-dimensional (2D) imaging generally adopts a 2-fold speed-up, and three-dimensional (3D) imaging generally adopts a 2 ⁇ 2-fold speed-up.
  • SENSE treats parallel reconstruction as a linear equation inversion problem in the image space, and when the sensitivity map is completely accurate, SENSE can produce an optimal solution in the least squares sense.
  • accurate sensitivity maps are difficult to obtain, and small errors in the sensitivity maps can lead to strong curling artifacts in reconstructed images.
  • vSENSE variable acceleration sensitivity encoding
  • the vSENSE method obtains an adjusted sensitivity map from a fully sampled or undersampled image frame by a lower factor, and then reconstructs other image frames undersampled with a higher acceleration factor.
  • vSENSE has a high acceleration capability, it has three distinct disadvantages.
  • the initial sensitivity map is obtained from an additional reference scan, which may introduce inconsistencies;
  • the SENSE image frame is 2 times faster It is considered to be accurate, and its reconstructed images still have potential artifacts; the highest speedup factor in the 3D vSENSE method is 8, and many artifacts have already appeared, and its speedup effect is still limited.
  • the purpose of the present invention is to solve the problems in the prior art that multi-frame imaging reconstruction cannot be self-calibrated, the reconstructed image has potential artifacts, and the speed-up effect is greatly limited, and provides a fast imaging method combining k-space and image-space reconstruction.
  • the English name is joint K-space and Image-space Parallel Imaging, which can be referred to as the KIPI method in the future
  • device which can recover accurate images from highly undersampled k-space data.
  • the present invention provides an imaging method combining k-space and image space reconstruction for reconstructing an undersampled image frame, the undersampled image frame comprising a first undersampled image frame with an automatic calibration signal and a second undersampled image frame without an automatic calibration signal and an acceleration factor not lower than 2, and the acceleration factor of the first undersampled image frame is not higher than the acceleration factor of the second undersampled image frame;
  • the reconstruction method steps are as follows:
  • S2 Divide the coil image of each channel by the channel merged image to obtain the coil sensitivity map of each channel; identify the support area from the channel merged image, and perform smooth denoising on the support area in the sensitivity map , extrapolate the unsupported area in the sensitivity map to get the optimized sensitivity map;
  • the parallel imaging method for self-calibration in k-space is the GRAPPA method.
  • the GRAPPA method is preferably GRAPPA with Tikhonov regularization.
  • the first under-sampled image frame and the second under-sampled image frame are both two-dimensional images; the acceleration factor of the first under-sampled image frame is preferably 2, and the second under-sampled image frame is preferably 2.
  • the acceleration factor of the frame is preferably 2 to 4.
  • the first under-sampled image frame and the second under-sampled image frame are both three-dimensional images; the acceleration factor of the first under-sampled image frame is preferably 2 ⁇ 2, and the second under-sampled image frame is preferably 2 ⁇ 2.
  • the acceleration factor of the image frame is preferably N ⁇ M, where 2 ⁇ N ⁇ 4, 2 ⁇ M ⁇ 4, and the total acceleration factor does not exceed 12.
  • all coil images are merged by a square root method or an adaptive method.
  • smoothing and denoising is performed on the support area in the sensitivity map by fitting.
  • the correction factor map needs to be filtered by a filter to remove abnormal values.
  • a truncated singular value regularization method is preferably used.
  • the present invention provides a magnetic resonance imaging apparatus, which includes a magnetic resonance scanner and a control unit, where a computer program is stored in the control unit, and when the computer program is executed, is used to realize the first
  • the multi-frame image reconstruction method according to any one of the aspects of the aspect; the magnetic resonance scanner is configured to acquire data of the first under-sampled image frame and the second under-sampled image frame.
  • the present invention has the following beneficial effects:
  • the present invention does not need the ACS data of all frames, and greatly improves the net acceleration effect.
  • the present invention is capable of generating accurate sensitivity maps using undersampled image frames with acceleration factors greater than or equal to 2. Since the first undersampled image frame is first reconstructed by a self-calibrating parallel imaging method in k-space, such as GRAPPA, the method enables self-calibration without the need for additionally acquired sensitivity maps.
  • the invention combines the robustness of the self-calibrated parallel imaging method in k-space, and can further speed up the acquisition speed compared with the original vSENSE method.
  • This method is especially suitable for 3D multi-frame imaging because it does not need to use fully sampled image frames.
  • the present invention allows the forward-looking acceleration factor in the direction of phase encoding and slice selection encoding to be increased by a factor of 12, and produces source images and final results that are consistent with the ground truth. target parameter image.
  • the images reconstructed by the present invention from highly undersampled data have no apparent aliasing artifacts.
  • Figure 1 is a graph of the resulting reconstructions in the Examples; comparison of reconstructed +6 ppm phantom images with GRAPPA and KIPI.
  • a is the sum of squares (RSS) reconstruction result of the fully sampled k-space
  • d is the Z spectrum and full k-space results (solid line) obtained by GRAPPA (dotted line) and KIPI (double dashed line) in the region of interest of the selected a image
  • e and f are the b image, c image and a, respectively Difference map between images.
  • Figure 2 is an amide proton transfer weighted (APTw) parameter image of the water model in the Examples.
  • APTw amide proton transfer weighted
  • Figure 3 is a GRAPPA and KIPI reconstructed +6ppm brain image in the Examples.
  • a is the RSS reconstruction of the fully sampled k-space
  • d is the selected source image of healthy volunteers Z-spectra and full k-space results (solid line) obtained by GRAPPA (dotted line) and KIPI (double-dashed line) in the region of interest of image a
  • Figure 4 is a 2D APTw parameter image of healthy volunteers in the Example.
  • a is the APTw image calculated from the full sampling data
  • Figure 5 is a -4ppm 3D-CEST source image of healthy volunteers in Examples (each group is 5 images selected from 72 images).
  • b is the source image of healthy volunteers reconstructed on undersampled data with variable acceleration factor using GRAPPA
  • c is the source image of healthy volunteers reconstructed using KIPI on variable acceleration factor
  • d, e are the difference maps of the GRAPPA and KIPI reconstruction results relative to the true value a, respectively, with reconstruction errors (RNMSE) of 0.026 and 0.014, respectively.
  • RPMSE reconstruction errors
  • FIG. 6 is an image of APTw parameters of healthy volunteers in Example (each group is 5 images selected from 72 images).
  • b is the APTw image of healthy volunteers reconstructed using GRAPPA on the undersampled data with variable acceleration factor
  • c is the APTw image reconstructed using KIPI in variable acceleration factor Reconstructed APTw images of healthy volunteers on undersampled data with acceleration factors.
  • Arrows indicate artifacts in variable acceleration factor GRAPPA APTw images.
  • FIG. 7 is an APTw image of healthy volunteers in Example (each group is 5 images selected from 72 images).
  • b is the APTw image of healthy volunteers reconstructed using GRAPPA on the undersampled data with variable acceleration factor
  • c is the APTw image reconstructed using KIPI in variable acceleration factor Reconstructed APTw images of healthy volunteers on undersampled data with acceleration factors.
  • Arrows indicate artifacts in variable acceleration factor GRAPPA APTw images.
  • FIG. 8 is a schematic flowchart of the KIPI method of the present invention, wherein the first undersampled image frame after reconstruction is a calibration frame.
  • the sensitivity map SE i of the ith coil can be obtained by dividing mi by the coil combination image ⁇ .
  • s N*1 represents the folded channel image vector
  • the superscripts 1 and 2 represent two different spatial aliasing positions.
  • Image vectors representing the two spatial aliasing locations, respectively.
  • ⁇ 1 and ⁇ 2 represent the pixel values of the two spatially aliased positions in the combined coil image, respectively.
  • Coil sensitivity vectors representing the two spatially aliased locations, respectively.
  • I is the identity matrix. and represent the estimates of ⁇ 1 and ⁇ 2 , respectively
  • Equation [2-4] actually illustrates that applying the sensitivity map derived from the GRAPPA reconstruction result to SENSE produces results that are consistent with the GRAPPA reconstruction and are consistent with SENSE
  • the acceleration factor is irrelevant.
  • m i is completely accurate, as long as the sensitivity used conforms to the definition of Equation [2], Equations [3] and [4] are valid. Since different image frames in the multi-frame imaging technique have the same sensitivity, the sensitivity map derived from the GRAPPA reconstructed frame can be applied to other frames.
  • the calibration signal (ACS) is denoted as the first under-sampled image frame; other frames are under-sampled with a higher acceleration factor, without ACS, denoted as the second under-sampled image frame.
  • the present invention performs GRAPPA reconstruction on the first under-sampled image frame with a lower acceleration factor as a calibration frame, then calculates the corresponding sensitivity map by using equation [2], and then applies this sensitivity map to perform SENSE reconstruction on the second under-sampled image frame .
  • the accuracy of SENSE reconstruction is only related to the sensitivity map used, not to the specific image contrast.
  • the above derivation has demonstrated that applying SENSE reconstruction at the calibration frame can achieve the same image quality as GRAPPA with a low acceleration factor. And since the calibration frame and the other second undersampled image frame have the same sensitivity (ideally accurate, but not available) and the imaged object is consistent, then use the above derived sensitivity map (actual, but available) on the other frame to perform SENSE reconstruction can also obtain images with similar quality to the calibration frame.
  • the flow chart of the imaging method combining k-space and image space reconstruction provided by the present invention is shown in FIG. 8 , and the specific implementation process thereof will be described below.
  • the multi-frame imaging reconstruction method is mainly used for reconstructing undersampled image frames acquired by a magnetic resonance scanner, wherein the undersampled image frames should include a first undersampled image frame with ACS and a second undersampled image without ACS frame.
  • the acceleration factor AF of the second under-sampled image frame should not be lower than 2, and the acceleration factor of the first under-sampled image frame should not be higher than the acceleration factor of the second under-sampled image frame, so as to save the cost of acquiring the second under-sampled image frame. required time.
  • the first undersampled image frame only needs one frame, and the second undersampled image frame has many frames, so if the calibration frame can be obtained using the first undersampled image frame, and then the rest of the second undersampled image frame is performed. Reconstruction can greatly improve the acquisition speed.
  • the undersampled image frame acquired by the magnetic resonance scanner may be either a 2D image or a 3D image.
  • the specific acceleration factor of the undersampled image frame needs to be adjusted according to the actual situation. If the undersampled image frame is a 2D image, the acceleration factor of the first undersampled image frame is preferably 2, and the acceleration factor of the second undersampled image frame is preferably 2. to 4. If the undersampled image frame is a 3D image, the acceleration factor of the first undersampled image frame is preferably 2 ⁇ 2, and the acceleration factor of the second undersampled image frame is preferably N ⁇ M, where 2 ⁇ N ⁇ 4, 2 ⁇ M ⁇ 4, and the total acceleration factor does not exceed 12.
  • the GRAPPA method belongs to the prior art, which reconstructs a complete image frame by calculating the coil weight and applying it to the under-sampled area. Further, if the GRAPPA method is used, the present invention recommends using GRAPPA with Tikhonov regularization.
  • all coil images can be combined by a square root (RSS) method, and of course, an adaptive combine method can also be used.
  • RSS square root
  • S2 Divide the coil image of each channel by the previous channel merged image to obtain the coil sensitivity map of each channel.
  • the support regions are identified from the aforementioned channel merged images, and the support regions in the sensitivity map are smoothed and denoised, and the non-support regions in the sensitivity map are extrapolated to obtain the optimized sensitivity map.
  • the support area in the sensitivity map can be smoothed and denoised by fitting.
  • the fitting includes local fitting and global fitting.
  • the smoothing and denoising of the support area can also be performed in the form of filter filtering.
  • a correction factor map is obtained based on the reference image and the retrospectively reconstructed image, and the pixel value of each position in the correction factor map is the quotient of the pixel value of the corresponding position in the reference image and the retrospectively reconstructed image, that is, the reference image and the retrospectively reconstructed image are each Pixel quotient results in a correction factor map.
  • SENSE reconstruction can be performed using the aforementioned optimized sensitivity map, and the reconstructed image after SENSE is multiplied by the aforementioned correction factor map, namely A complete image frame with reduced artifacts is available.
  • multiplying the two images refers to multiplying the pixel values at the same position in the two images point by point, as the corresponding image in the complete image frame. The pixel value of the location.
  • the second under-sampled image frame in the present invention has multiple frames, but different second under-sampled image frames may have different acceleration factors.
  • the correction factor map used by it also needs to be obtained based on the same acceleration factor.
  • the calibration frame needs to be retrospectively undersampled based on the acceleration factor X, and then the retrospectively reconstructed image is reconstructed through SENSE, and based on the reference
  • the correction factor map is obtained from the image and the retrospectively reconstructed image; when the acceleration factor of the second undersampled image frame of another frame is Y, then in S3, the calibration frame needs to be retrospectively undersampled based on the acceleration factor Y, and then passed through SENSE
  • the retrospectively reconstructed image is reconstructed, and a correction factor map is obtained based on the reference image and the retrospectively reconstructed image.
  • the above S1 to S4 constitute the imaging method (KIPI) of the present invention that combines k-space and image space reconstruction.
  • the above-mentioned KIPI method can be integrated into a control unit of a magnetic resonance imaging device, thereby forming a magnetic resonance imaging device capable of automatically performing multi-frame imaging reconstruction.
  • the magnetic resonance imaging apparatus includes a magnetic resonance scanner and a control unit, wherein a computer program is stored in the control unit, and when the computer program is executed, the above-mentioned KIPI method can be implemented.
  • the under-sampled image data (the first under-sampled image frame and the second under-sampled image frame) required by the KIPI method are acquired by the magnetic resonance scanner in advance.
  • the above-mentioned magnetic resonance scanner can be realized by using the existing technology, and is a mature commercial product, and will not be described again.
  • control unit can be a general-purpose processor, including a central processing unit (Central Processing Unit, CPU), a network processor (Network Processor, NP), etc.; it can also be a digital signal processor (Digital Signal Processing, DSP), application-specific integrated circuits (Application Specific Integrated Circuit, ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.
  • CPU Central Processing Unit
  • NP Network Processor
  • DSP Digital Signal Processing
  • ASIC Application Specific Integrated Circuit
  • FPGA Field-Programmable Gate Array
  • control unit in addition to the programs for realizing the KIPI method, the control unit should also have imaging sequences and other software programs necessary for realizing multi-frame imaging.
  • the imaging method and device combining k-space and image space reconstruction of the present invention can be applied to any multi-frame imaging technology in magnetic resonance imaging that requires multiple acquisitions of the same imaging object, including but not limited to relaxation times (T1, T2 ) imaging, diffusion imaging, perfusion imaging, functional imaging (fMRI), magnetic resonance spectroscopy (MRS) imaging, chemical exchange saturation transfer (CEST) imaging.
  • T1, T2 relaxation times
  • fMRI functional imaging
  • MRS magnetic resonance spectroscopy
  • CEST chemical exchange saturation transfer
  • the phantom consists of a flask filled with 2% agarose gel and two test tubes. One tube was filled with 10% bovine serum albumin (BSA), dissolved in phosphate buffered saline (PBS), and the other was filled with 5% bovine serum albumin, also dissolved in PBS.
  • BSA bovine serum albumin
  • PBS phosphate buffered saline
  • the sequence used was the CEST imaging sequence and the MRI parameter measured was the amide proton transfer effect magnitude.
  • the signal mean per saturation frame (NSA) was 2.
  • the 2D human study used the same parameters as the phantom study.
  • ACS adopts embedded acquisition, and the matrix size is 24 ⁇ 76 ⁇ 72.
  • a total of 7 CEST saturation-shifted frames were acquired for amide proton transfer-weighted (APTw) imaging, including (S0), ⁇ 3, ⁇ 3.5, and ⁇ 4-ppm.
  • the GRE sequence with the same field of view, orientation and resolution as the CEST sequence was used in the 2D and 3D experiments, with a TR of 30 ms.
  • the GRE sequence was run in dual-echo mode with TE of 4.92ms and 9.84ms, respectively.
  • the first undersampled image frame was reconstructed using GRAPPA with Tikhonov regularization and filled with ACS data as the calibration frame.
  • Fourier transform is performed on the k-space data of the reconstructed calibration frame to obtain the coil image of each channel, and then the coil images of all channels are reconstructed and merged by the square root (RSS).
  • the 2D GRAPPA kernel is 4 ⁇ 5, representing 4 acquired phase-encoded lines and 5 frequency-encoded points, that is, a missing point is fitted using 20 points in each channel.
  • the 3D GRAPPA kernel is 4 ⁇ 5 ⁇ 4 (respectively, phase encoding ⁇ frequency encoding ⁇ layer selection encoding direction).
  • the original sensitivity profile for each channel is calculated from the reconstructed calibration frame image by dividing the coil image for each channel by the above RSS image.
  • Sensitivity maps computed in this way have the same geometry as the machine-scanned image, so no registration is required. Therefore thresholding was used to identify support regions from RSS images (the threshold was set around 0.1), and morphological imaging was used to fill holes in the sensitivity maps and smooth the region boundaries.
  • a local weighted polynomial regression (LWPR) fitting with a cubic weighted kernel is used to smooth and denoise the support region in the sensitivity map, and extrapolate the non-support region in the sensitivity map to obtain an optimized sensitivity map.
  • the polynomial used in this embodiment is a second-order polynomial, the window width of the support area is 12, and the window width of the non-support area is 24.
  • the correction factor map is defined as the point-by-point division of the aliased image by the retrospectively reconstructed image, that is, the pixel value of each position in the correction factor map is the quotient of the pixel value of the reference image and the corresponding position in the retrospectively reconstructed image.
  • the correction factor map in this embodiment needs to be filtered by a median filter with a window of 3 ⁇ 3 to remove outliers.
  • the calculation formula of the correction factor graph C can be expressed as follows:
  • This step uses the truncated singular value regularization method to perform SENSE reconstruction on the second undersampled image frame, and discards singular values less than 2% of the maximum value. Then, using the artifact suppression method, the image generated by the SENSE method is multiplied point by point with the correction factor map to further reduce the error, and a complete image frame with suppressed artifacts can be obtained.
  • Perform retrospective undersampling to generate two different correction factor maps the corresponding correction factor maps corresponding to the second undersampled image frame SENSE reconstruction in step 4.
  • the APTw parameter image is calculated as follows. First, the source image is registered to the first undersampled frame 3.5ppm. Second, the phase difference of the GRE images acquired by two different TEs was calculated as a B0 map. Third, corrected +3.5-ppm and -3.5-ppm signal values were generated for each voxel from the calculated B0 map. Finally, APTw parametric images are obtained by subtracting the corrected 3.5-ppm and +3.5-ppm images.
  • the GRAPPA method (Fig. 1b,e) has much larger errors (arrows) than the KIPI method (Fig. 1c,f).
  • RNMSE reconstruction error
  • the z-spectra of conventional GRAPPA at the circled place (dashed line in Fig. 1d) produces a large error compared to the true value (solid line in Fig. 1d). While the results produced by KIPI (double-dashed lines in Fig. 1d) are almost identical to the true values.
  • Figure 2 presents an image of APTw parameters computed from the source image ( Figure 1).
  • the APTw images generated by the KIPI method (Fig. 2c) are in good agreement with the ground truth (fully sampled, Fig. 2a), showing only slight differences.
  • Figure 3 shows images of healthy human brains under the same experimental conditions. Similar to the phantom study, the results also validate that the reconstruction of KIPI is more accurate than that of GRAPPA.
  • KIPI-generated source images (Fig. 3c) agree better with ground truth (Fig. 3a) than source images reconstructed from the same data using GRAPPA (Fig. 3b; reconstruction errors are 0.018 and 0.029, respectively).
  • the z-spectrum produced by the KIPI method (Fig. 3d; double-dashed line) is almost indistinguishable from the full k-space spectrum (Fig. 3d; solid line), while GRAPPA causes significant errors (Fig. 3d; dashed line).
  • the source images generated by KIPI (Fig. 5c) are more in line with the ground truth than GRAPPA (Fig. 5b) reconstructions, despite using the same variable acceleration factor undersampled data.
  • aliasing artifacts were also evident in the GRAPPA images (Fig. 5b,d; arrows).
  • FIG. 6 is an APTw parameter image generated from the source image of FIG. 5 after B0 correction and image registration.
  • a large number of artifacts can be seen in the APTw image reconstructed by the conventional GRAPPA method (Fig. 6b), mainly manifested as aliasing artifacts in the slice selection direction.
  • the APTw images generated by the KIPI method (Fig. 6c) are almost indistinguishable from the ground truth (Fig. 6a).
  • Figure 7 shows images of APTw parameters in healthy volunteers with higher acceleration factors.
  • the folding artifacts of GRAPPA are mainly manifested in the phase-encoding direction.
  • the APTw images reconstructed with GRAPPA have obvious artifacts, which are basically absent from the KIPI results.
  • KIPI acceleration factor of up to 12 for a single source image frame. Overall, the net effective acceleration factor can reach 8.

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Radiology & Medical Imaging (AREA)
  • Health & Medical Sciences (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Signal Processing (AREA)
  • General Health & Medical Sciences (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Condensed Matter Physics & Semiconductors (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Magnetic Resonance Imaging Apparatus (AREA)

Abstract

一种结合k空间和图像空间重建的成像方法和装置，属于磁共振成像领域。该方法首先使用k空间并行成像方法对被测对象在磁共振参数测量所需的多帧成像中得到的带有自动校准信号的欠采样图像帧进行重建；然后针对其他没有自动校准信号的欠采样图像帧数据，在图像空间进行重建。该方法从欠采样数据中生成准确的SENSE方法所需要的灵敏度图，而无需额外采集。该方法产生与传统GRAPPA方法几乎一致的图像，并且没有明显伪影，速度却是传统GRAPPA方法的四倍及以上，尤其适合用来加速三维(3D)多帧磁共振成像。

Description

结合k空间和图像空间重建的成像方法和装置 技术领域
本发明属于磁共振成像领域，可以针对需要多帧图像数据的磁共振参数测量实现快速采集。
背景技术
磁共振成像(MRI)可以用来测量包括核自旋密度、纵向弛豫时间(T1)，横向弛豫时间(T2)，表观扩散系数(ADC)，脑血流速率，磁化转移速率等组织参数。这些参数的信号来源可以是生物体内存在的内源性化合物的核(如水中的 1H)，也可以是外源性物质或示踪剂。组织的MRI参数的直接量化可以提供有关神经、肌肉骨骼、肝脏和心肌成像中各种病理上的更有价值的临床信息。
测量这些参数的成像技术包括弛豫时间成像、扩散成像、灌注成像、功能磁共振(fMRI)和化学交换饱和转移(CEST)成像，它们大都需要对同一个成像对象采集具有脉冲序列参数调制的多帧图像数据集。这些可调制的脉冲序列参数包括回波时间(TE)、翻转角度或梯度强度等，例如扩散成像需要对同一个成像对象采集不同梯度磁场下的数据。目标参数估计可以通过将多帧图像数据集中的信号演化与数学模型进行拟合来实现。由于需要多次采集，扫描时间很长，这限制了其在临床中的应用；另一方面较长的扫描时间容易引入额外的运动，这使得技术本身对运动比较敏感。因此减少数据采集时间，加速成像过程不仅方便患者，也有利于提高图像质量。
为了加速多帧成像中的参数测量，国内外学者已经提出了很多方法，以CEST技术为例有Keyhole-CEST，SLAM，k-w ROSA，CS-CEST等方法。Keyhole-CEST将获取的低分辨率图像以及高分辨率的参考图像相结合以减少每次所需采集的数据，但是其牺牲了图像的高频信息。SLAM方法使用从侦察扫描中获得的先验定位知识直接从任意形状的感兴趣区域产生区域CEST测量值，具有比传统单体素和多体素方法高得多的信噪比效率，但是区域内部组织异质性的信息丢失了。基于模型的方法，k-w ROSA通过将基于子空间的Z谱信号分解合并到测量模型 中，可以直接从完整或不完整的测量中估计感兴趣的不对称Z谱，但是其加速能力有限。基于压缩感知的CS-CEST方法利用图像在变换域的稀疏性进行提速，但是在实际应用中尚未实现完全的随机欠采样。
并行成像方法因其实用性和鲁棒性而被临床广泛使用。目前可大致分为两大类，以GRAPPA为代表的自校准的k空间方法，和以SENSE为代表的基于显性线圈灵敏度的图像空间方法。GRAPPA是一种自校准的k空间并行成像算法，其通过多通道数据的线性组合来拟合缺失点。GRAPPA的校准方程如下：
y＝Aw            [1]
其中A代表由自动校准信号(ACS)生成的源数据矩阵，y表示目标数据向量，w代表要拟合的权重。过去的研究表明当加速因子较低时，GRAPPA可以产生准确的图像，并具有较高的鲁棒性。对于需要多帧数据的参数测量来说，由于需要采集很多帧图像，传统的GRAPPA方法需要每个图像帧都采集ACS，这大大降低了提速效果；如果只采集一个ACS，将从其中计算出的权值应用到所有欠采样帧，又会引入未知的伪影。另一方面，当加速因子很高时，GRAPPA只能从相距较远的k空间采集数据来拟合缺失数据，这增加了误差。GRAPPA的原理限制了其提速能力，实际中二维(2D)成像一般采用2倍提速，三维(3D)成像一般采用2×2倍提速。
SENSE将并行重建视为图像空间的线性方程求逆问题，在灵敏度图完全准确的情况下，SENSE可以产生最小二乘意义上的最优解。但是实际中，准确的灵敏度图很难获得，而且灵敏度图中微小的误差可以导致重建图像中很强的卷折伪影。最近，一种变加速敏感度编码(vSENSE)方法被提出，来加速多帧图像数据采集和参数测量。由于在多帧成像中，不同帧的图像是从相同成像对象获得的，理论上具有相同的线圈灵敏度。基于这一假设，vSENSE方法从一个全采样或者较低倍数欠采样的图像帧中获得调整的灵敏度图，然后去重建其他以较高加速因子欠采样的图像帧。尽管vSENSE具有较高的加速能力，但它有三个明显的缺点。首先，其不是自校准的方法，从额外的参考扫描中获取初始灵敏度图，可能会引入不一致性；其次，在最新的3D vSENSE方法中，由于全采样图像不可获得，提速2倍的SENSE图像帧被视为是准确的，其重建的图像仍具有潜在的伪影；3D vSENSE方法中最高加速倍数为8，而且已经出现较多伪影，其提速效果仍受 到限制。
因此，在需要多帧成像的MR参数测量领域，提出一种能够自校准重建，同时允许更高加速因子且具有高鲁棒性的方法具有非常重要的意义。
发明内容
本发明的目的在于解决现有技术中多帧成像重建无法自校准，而且重建图像具有潜在伪影、提速效果受到较大限制等问题，并提供一种结合k空间和图像空间重建的快速成像方法(英文名为joint K-space and Image-space Parallel Imaging，后续可简称为KIPI方法)和装置，可以从高度欠采样的k空间数据中恢复精确的图像。
本发明所采用的具体技术方案如下：
第一方面，本发明提供了一种结合k空间和图像空间重建的成像方法，用于对欠采样图像帧进行重建，所述欠采样图像帧包含带有自动校准信号的第一欠采样图像帧和没有自动校准信号且加速因子不低于2的第二欠采样图像帧，且第一欠采样图像帧的加速因子不高于第二欠采样图像帧的加速因子；
所述重建方法步骤如下：
S1：针对第一欠采样图像帧，利用k空间中自校准的并行成像方法重建得到完整图像帧作为校准帧；对校准帧的k空间数据进行傅里叶变换后得到每个通道的线圈图像，合并所有通道的线圈图像后得到通道合并图像；
S2：将每个通道的线圈图像除以所述通道合并图像，得到每个通道的线圈灵敏度图；从所述通道合并图像中识别出支撑区域，并对灵敏度图中的支撑区域进行平滑去噪，对灵敏度图中的非支撑区域进行外推，得到优化的灵敏度图；
S3：利用优化的灵敏度图，在所述校准帧上进行加速因子为1的SENSE重建，生成参考图像；同时，对所述校准帧进行回顾性欠采样且回顾性欠采样的加速因子与所述第二欠采样图像帧的加速因子相同，再对欠采样后的数据进行SENSE重建，生成具有潜在混叠伪影的回顾性重建图像；基于参考图像和回顾性重建图像得到校正因子图，且校正因子图中每个位置的像素值为参考图像和回顾性重建图像中对应位置像素值的商；
S4：针对第一欠采样图像帧之外的其余第二欠采样图像帧，利用所述优化的灵敏度图进行SENSE重建，将SENSE重建后的图像与所述校正因子图相乘， 得到抑制伪影的完整图像帧。
作为第一方面的优选，所述S1中，所述k空间中自校准的并行成像方法为GRAPPA方法，通过计算出线圈权重并应用于欠采样区域，重建得到完整图像帧。
进一步的，所述GRAPPA方法优选为带有Tikhonov正则化的GRAPPA。
作为第一方面的优选，所述第一欠采样图像帧和第二欠采样图像帧均为二维图像；所述第一欠采样图像帧的加速因子优选为2，所述第二欠采样图像帧的加速因子优选为2到4。
作为第一方面的优选，所述第一欠采样图像帧和第二欠采样图像帧均为三维图像；所述第一欠采样图像帧的加速因子优选为2×2，所述第二欠采样图像帧的加速因子优选为N×M，其中2≤N≤4，2≤M≤4，且总加速因子不超过12。
作为第一方面的优选，所述S1中，通过平方根方法或者自适应方法合并所有的线圈图像。
作为第一方面的优选，所述S2中，对灵敏度图中的支撑区域通过拟合方式进行平滑去噪。
作为第一方面的优选，所述S3中，所述校正因子图需经过滤波器滤波，以去除异常值。
作为第一方面的优选，所述S4中，利用所述优化的灵敏度图进行SENSE重建时，优选采用截断奇异值正则化方法。
第二方面，本发明提供了一种磁共振成像设备，其包括磁共振扫描器以及控制单元，所述控制单元中存储有计算机程序，当所述计算机程序被执行时，用于实现如第一方面中任一项方案所述的多帧图像重建方法；所述磁共振扫描器用于获取所述第一欠采样图像帧和第二欠采样图像帧数据。
本发明相对于现有技术而言，具有以下有益效果：
本发明无需所有帧的ACS数据，大大提高了净加速效果。本发明能够使用加速因子大于等于2的欠采样图像帧来生成准确的灵敏度图。由于第一欠采样图像帧首先被k空间中自校准的并行成像方法(如GRAPPA)重建，因此不需要额外采集的灵敏度图，该方法能够实现自校准。本发明结合了k空间中自校准的并行成像方法的鲁棒性，相比于原来的vSENSE方法可进一步提速采集速度。
由于无需使用全采样图像帧，该方法尤其适用于3D多帧成像。而且，在获 取参数测量所需要的源图像时，本发明允许在相位编码和层选编码方向上的前瞻性加速因子提高至原来的12倍，并且产生与真值结果相一致的源图像和最终的目标参数图像。另一方面，本发明从高度欠采样数据中重建的图像没有明显的混叠伪影。
附图说明
图1为实施例中的结果重建图；用GRAPPA和KIPI重建+6ppm体模图像的比较。其中a为全采样k空间的平方和(RSS)重建结果；b为当加速因子(AF)＝4时，GRAPPA重建的体模源图像；c为当AF＝4时，KIPI重建的体模源图像；d为选定a图像的感兴趣区域中GRAPPA(虚线)和KIPI(双划线)得到的Z谱与全k空间结果(实线)；e和f分别为b图像、c图像与a图像之间的差分图。且b和c图像使用相同的变加速因子欠采样数据，选取第一个+3.5ppm作为AF＝2的第一欠采样图像帧，选取S0和3.5ppm(不包括第一欠采样图像帧)作为AF＝2，以及其他46帧AF＝4。
图2为实施例中的水模的酰胺质子转移加权(APTw)参数图像。其中a为全采样数据计算的APTw图像；b为GRAPPA利用从第一欠采样图像帧ACS数据中获得的权值重建其他第二欠采样图像帧得到的APTw图像，产生明显的伪影；c为KIPI利用从校准帧导出的灵敏度图重建其他第二欠采样图像得到的APTw图像。且b和c使用相同的变加速因子欠采样数据，选取第一个+3.5ppm作为AF＝2的第一欠采样图像帧，选取S0和±3.5ppm(不包括第一欠采样图像帧)作为AF＝2，以及其他46帧AF＝4。
图3为实施例中的GRAPPA和KIPI重建+6ppm脑图像。其中a为全采样k空间的RSS重建；b为当AF＝4时，GRAPPA重建的健康志愿者的源图像；c为当AF＝4时，KIPI重建的健康志愿者的源图像；d为选定a图像的感兴趣区域中GRAPPA(虚线)和KIPI(双划线)得到的Z谱与全k空间结果(实线)；e和f分别b、c图像与a图像之间的差分图。且b和c图像使用相同的变加速因子欠采样数据，选取第一个+3.5ppm作为AF＝2的第一欠采样图像帧，选取S0和±3.5ppm(不包括第一欠采样图像帧)作为AF＝2，以及其他46帧AF＝4。
图4为实施例中健康志愿者的2D APTw参数图像。其中a为全采样数据计算的APTw图像；b为AF＝4时，GRAPPA重建的APTw图像，重建其他帧使用 校准帧ACS数据中获得的权值；c为AF＝4时，KIPI重建的APTw图像，在SENSE步骤时使用从校准帧导出的灵敏度图。且b和c图像使用相同的变加速因子欠采样数据，选取第一个+3.5ppm作为AF＝2的第一欠采样图像帧，选取S0和±3.5ppm(不包括第一欠采样图像帧)作为AF＝2，以及其他46帧AF＝4。
图5为实施例中健康志愿者的-4ppm 3D-CEST源图像(每组为从72张图中挑出的5张图)。其中a为AF＝2×1的传统GRAPPA重建获得的-4ppm的源图像；b为使用GRAPPA在变加速因子的欠采样数据上重建的健康志愿者源图像；c为使用KIPI在变加速因子的欠采样数据上重建的健康志愿者源图像；d，e分别为GRAPPA和KIPI重建结果相对于真实值a的差别图，重建误差(RNMSE)分别为0.026和0.014。箭头指向的是GRAPPA重建的混叠伪影。其中变提速欠采样数据为，+3.5ppm帧有ACS数据，AF＝2×2为第一欠采样图像帧；其他6帧没有ACS数据且AF＝2×4为第二欠采样图像帧。
图6为实施例中健康志愿者的APTw参数图像(每组为从72张图中挑出的5张图)。其中a为所有帧的AF＝2×1数据和ACS数据的GRAPPA重建得到的APTw图像；b为使用GRAPPA在变加速因子的欠采样数据上重建的健康志愿者APTw图像；c为使用KIPI在变加速因子的欠采样数据上重建的健康志愿者APTw图像。箭头指示变加速因子GRAPPA APTw图像中的伪影。其中+3.5ppm帧有ACS数据，AF＝2×2为第一欠采样图像帧；其他6帧没有ACS数据且AF＝2×4为第二欠采样图像帧。
图7为实施例中健康志愿者的APTw图像(每组为从72张图中挑出的5张图)。其中a为所有帧的AF＝2×1数据和ACS数据的GRAPPA重建得到的APTw图像；b为使用GRAPPA在变加速因子的欠采样数据上重建的健康志愿者APTw图像；c为使用KIPI在变加速因子的欠采样数据上重建的健康志愿者APTw图像。箭头指示变加速因子GRAPPA APTw图像中的伪影。其中+3.5ppm帧有ACS数据，AF＝2×2为第一欠采样图像帧；其他6帧没有ACS数据且AF＝4×3为第二欠采样图像帧。
图8为本发明KIPI方法的流程示意图，其中完成重建的第一欠采样图像帧即为校准帧。
具体实施方式
下面结合附图和具体实施方式对本发明做进一步阐述和说明。本发明中各个实施方式的技术特征在没有相互冲突的前提下，均可进行相应组合。
对于SENSE重建方法，如果第i(1≤i≤N)个线圈图像m i已知，那么第i个线圈的灵敏度图SE i可以通过将m i除以线圈组合图像ρ得到。
SE i＝m i/ρ          [2]
因此，对于加速因子(AF)为2(不失一般性)回顾性的SENSE重建，有以下方程：
Figure PCTCN2021140280-appb-000001
其中s N*1代表卷折后的通道图像向量，上标1和2代表不同的两个空间混叠位置。
Figure PCTCN2021140280-appb-000002
和
Figure PCTCN2021140280-appb-000003
分别代表两个空间混叠位置的图像向量。ρ 1和ρ 2分别代表线圈组合图像中两个空间混叠位置的像素值。
Figure PCTCN2021140280-appb-000004
和
Figure PCTCN2021140280-appb-000005
分别代表两个空间混叠位置的线圈灵敏度向量。
等式[3]的最小二乘解如下：
Figure PCTCN2021140280-appb-000006
其中
Figure PCTCN2021140280-appb-000007
I是单位矩阵。
Figure PCTCN2021140280-appb-000008
和
Figure PCTCN2021140280-appb-000009
分别代表ρ 1和ρ 2的估
计值。
在实际情况中，通道图像通常是未知的。注意到，如果把GRAPPA重建的结果作为m i代入，等式[2-4]实际上阐明了把从GRAPPA重建结果导出的灵敏度图应用到SENSE，可以产生与GRAPPA重建一致的结果，并且与SENSE的加速因子无关。其实不论m i是否是完全准确的，只要使用的灵敏度符合等式[2]的定义，等式[3]和[4]都是成立的。由于多帧成像技术中不同图像帧具有相同的灵敏度，因此可以把从GRAPPA重建帧中导出灵敏度图应用到别的帧。
基于上述原理，在采样时本发明选择多帧成像中一个图像帧以较低的加速因子(例如2D成像是AF＝2，3D成像是AF＝2×2)进行欠采样而且保留中心部分的自动校准信号(ACS)，记为第一欠采样图像帧；对于其他帧以较高的加速因子进行欠采样，没有ACS，记为第二欠采样图像帧。本发明对较低加速因子的 第一欠采样图像帧进行GRAPPA重建作为校准帧，然后用等式[2]计算出相应的灵敏度图，再应用此灵敏度图对第二欠采样图像帧进行SENSE重建。在一般意义上，SENSE重建的准确度只与所使用灵敏度图的相关，而和具体的图像对比度无关。上述推导已经证明了在校准帧应用SENSE重建可获得与低加速因子的GRAPPA相同的图像质量。而由于校准帧和其他第二欠采样图像帧具有相同灵敏度(理想的准确的，但不可获得的)且成像对象一致，那么在其他帧使用上述导出灵敏度图(实际的，但可获得的)进行SENSE重建也可获得具有与校准帧相近质量的图像。本发明提供的结合k空间和图像空间重建的成像方法流程如图8所示，下面对其具体实现过程进行展开描述。
该多帧成像重建方法主要用于对磁共振扫描器获取到的欠采样图像帧进行重建，其中欠采样图像帧应当包含带有ACS的第一欠采样图像帧和没有ACS的第二欠采样图像帧。第二欠采样图像帧的加速因子AF不低于2，而且第一欠采样图像帧的加速因子应当不高于第二欠采样图像帧的加速因子，以此节省获取第二欠采样图像帧的所需时间。第一欠采样图像帧仅需一帧，而第二欠采样图像帧有许多帧，因此如果可以利用第一欠采样图像帧获得校准帧，进而对其余的每一帧第二欠采样图像帧进行重建，就可以大大提高采集速度。
需说明的是，磁共振扫描器获取到的欠采样图像帧既可以为2D图像，也可以是3D图像。欠采样图像帧的具体加速因子需要根据实际情况进行调整，如果欠采样图像帧为2D图像，那么第一欠采样图像帧的加速因子优选为2，第二欠采样图像帧的加速因子优选为2到4。如果欠采样图像帧为3D图像，那么第一欠采样图像帧的加速因子优选为2×2，第二欠采样图像帧的加速因子优选为N×M，其中2≤N≤4，2≤M≤4，且总加速因子不超过12。
该结合k空间和图像空间重建的成像方法，具体步骤如下：
S1：针对第一欠采样图像帧，利用k空间中自校准的并行成像方法GRAPPA重建得到完整图像帧作为校准帧。然后，对校准帧的k空间数据进行傅里叶变换后得到每个通道的线圈图像，合并所有通道的线圈图像后得到通道合并图像。
GRAPPA方法属于现有技术，其通过计算出线圈权重并应用于欠采样区域，重建得到完整图像帧。进一步的，若采用GRAPPA方法，本发明推荐采用带有Tikhonov正则化的GRAPPA。
但需要说明的是，在本步骤中，虽然推荐使用GRAPPA来实现第一欠采样图像帧的重建，但是根据情况也可以采取SPIRiT，CAIPIRINHA等其他的k空间中自校准的并行成像方法。
另外，在本步骤中，所有的线圈图像合并可以通过平方根(RSS)方法实现，当然也可以采用自适应方法合并(adaptive combine)方法实现。
S2：将每个通道的线圈图像除以前述通道合并图像，得到每个通道的线圈灵敏度图。从前述通道合并图像中识别出支撑区域，并对灵敏度图中的支撑区域进行平滑去噪，对灵敏度图中的非支撑区域进行外推，得到优化的灵敏度图。
在本步骤中，对灵敏度图中的支撑区域可以通过拟合方式进行平滑去噪。拟合包括局部拟合和全局拟合两种方式，当然也可以通过滤波器滤波的形式进行支撑区域的平滑去噪。
S3：利用优化的灵敏度图，在前述校准帧上进行加速因子为1的SENSE重建，生成参考图像。同时，对前述校准帧进行回顾性欠采样且回顾性欠采样的加速因子与前述第二欠采样图像帧的加速因子相同，再对欠采样后的数据进行SENSE重建，生成具有潜在混叠伪影的回顾性重建图像。然后基于参考图像和回顾性重建图像得到校正因子图，且校正因子图中每个位置的像素值为参考图像和回顾性重建图像中对应位置像素值的商，即参考图像和回顾性重建图像逐像素求商得到校正因子图。
需要注意的是，直接得到的校正因子图中可能存在异常值，因此最好需要将其经过滤波器滤波，去除异常值后再进行下一步。
S4：针对第一欠采样图像帧之外的其余每一帧第二欠采样图像帧，即可利用前述优化的灵敏度图进行SENSE重建，将SENSE重建后的图像与前述校正因子图相乘，即可得到抑制伪影的完整图像帧。
需说明的是，本步骤中将SENSE重建后的图像与校正因子图相乘时，两张图像相乘是指将两张图像中相同位置的像素值逐点相乘，作为完整图像帧中对应位置的像素值。
需要注意的是，本发明中的第二欠采样图像帧具有多帧，但是不同的第二欠采样图像帧可以具有不同的加速因子。在对每一帧第二欠采样图像帧进行SENSE重建时，其采用的校正因子图也需要基于相同的加速因子获得的。具体 而言，如果一帧第二欠采样图像帧的加速因子为X，那么在S3中需要也基于加速因子X对校准帧进行回顾性欠采样，然后通过SENSE重建回顾性重建图像，并基于参考图像和回顾性重建图像得到校正因子图；当另外一帧第二欠采样图像帧的加速因子为Y时，那么在S3中需要也基于加速因子Y对校准帧进行回顾性欠采样，然后通过SENSE重建回顾性重建图像，并基于参考图像和回顾性重建图像得到校正因子图。
在本步骤中，利用优化的灵敏度图进行SENSE重建时，优选采用截断奇异值正则化方法，丢弃部分较小奇异值。
由此，上述S1～S4即构成了本发明的结合k空间和图像空间重建的成像方法(KIPI)。在实际应用中，可以将上述KIPI方法集成于磁共振成像设备的控制单元中，进而形成一种能够自动进行多帧成像重建的磁共振成像设备。该磁共振成像设备包括磁共振扫描器以及控制单元，其中控制单元中存储有计算机程序，当该计算机程序被执行时，可以实现上述KIPI方法。而该KIPI方法所需的欠采样图像数据(第一欠采样图像帧和第二欠采样图像帧)则预先由磁共振扫描器获取。
上述磁共振扫描器可采用现有技术实现，属于成熟商用产品，不再赘述。
上述控制单元可以是通用处理器，包括中央处理器(Central Processing Unit，CPU)、网络处理器(Network Processor，NP)等；还可以是数字信号处理器(Digital Signal Processing，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现场可编程门阵列(Field-Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件。
需要说明的是，控制单元中除了存储有实现KIPI方法的程序之外，还应当具有实现多帧成像所必要的成像序列以及其他软件程序。
本发明的结合k空间和图像空间重建的成像方法和装置，可适用于磁共振成像中任何需要对同一成像对象进行多次采集的多帧成像技术，包括但不限于弛豫时间(T1，T2)成像，扩散成像，灌注成像，功能成像(fMRI)，磁共振波谱(MRS)成像，化学交换饱和转移(CEST)成像。
下面将基于CEST成像，通过一个实施例来进一步展示本发明前述KIPI方法所能够实现的技术效果，以便本领域技术人员更好地理解本发明的实质。
实施例
1、MRI实验
所有体模实验和人体实验均在一台3特斯拉(T)西门子扫描仪(MAGNETOM Prisma,Siemens Healthcare,Erlangen,Germany)上进行，使用64通道接收头线圈。体模由一个装满2％琼脂糖凝胶的烧瓶和两个试管组成。一管填充10％牛血清白蛋白(BSA)，溶于磷酸盐缓冲盐水(PBS)，另一管填充5％牛血清白蛋白也溶于PBS。人体研究得到了当地机构审查委员会的批准。
所使用的序列为CEST成像序列，测量的MRI参数为酰胺质子转移效应幅度。
对于体模,CEST扫描使用时长1.0s,2μT饱和脉冲，然后使用脂肪抑制的轴向2D多自旋回波(TSE)序列读出，采集参数为TE＝6.7ms；TR＝3s；FA＝90；FOV＝212×186mm 2；分辨率＝2.2×2.2mm 2；片厚＝5mm；采集矩阵大小＝96×96；涡轮系数＝42。一共采集了51个不同频率偏移的帧，包括不饱和帧S0和6～-6ppm的饱和帧，其中步长为0.5ppm。每个饱和帧的信号平均数(NSA)为2。2D人体研究使用了与体模研究相同的参数。
对于3D人体脑实验，使用SPACE-CEST序列进行采集数据，运行参数如下:TE＝17ms,TR＝3s,FOV＝212×212×201mm 3，分辨率＝2.8×2.8×2.8mm 3，采集矩阵大小＝76×76×72，涡轮系数＝140,NSA＝1.2,GRAPPA加速因子＝2×1(分别在相位编码和层选编码方向)。ACS采用嵌入式采集，矩阵大小为24×76×72。共采集7个CEST饱和偏移帧用于酰胺质子转移加权(APTw)成像，包括(S0),±3,±3.5,and±4-ppm。
对于B0校正，在2D和3D实验中采用与CEST序列相同视场、方位和分辨率的GRE序列，TR为30ms。GRE序列在双回波模式下运行，TE分别为4.92ms和9.84ms。
2、图像重建与分析
所有的处理和分析都是使用在PC电脑(3.2GHz)上编写的MATLAB(MathWorks,Natick,MA)软件离线进行的。
在2D实验中，选择第一个+3.5ppm作为第一欠采样图像帧，AF＝2。第二欠采样图像帧如下：S0和±3.5ppm(不包括第一欠采样图像帧)欠采样AF＝2，其 他46帧AF＝4。对于3D实验，选择+3.5ppm帧作为第一欠采样图像帧，AF＝2×2，其余6帧为第二欠采样图像帧且AF＝2×4(第一个3D实验)或AF＝4×3(第二个3D实验)；对于第一欠采样图像帧，将ACS矩阵尺寸从24×76×72减小到24×76×24，这意味着只保留了原有ACS的一部分。注意，不论2D还是3D实验，只有第一欠采样图像帧保留了ACS数据。
本实施例的KIPI分为以下几步：
1.首先，使用GRAPPA重建第一欠采样图像帧。使用带有Tikhonov正则化的GRAPPA重建第一欠采样图像帧，并将ACS数据填入作为校准帧。将重建好的校准帧的k空间数据进行傅里叶变换得到每个通道的线圈图像，然后通过平方根(RSS)重建合并所有通道的线圈图像，合并后得到的通道合并图像记为RSS图像。其中2D GRAPPA内核为4×5，代表4条采集到的相位编码线和5个频率编码点，也就是说，一个缺失点会使用每个通道中的20个点来拟合。在相同的意义下，3D的GRAPPA内核为4×5×4(分别为相位编码×频率编码×层选编码方向)。
2.其次，计算线圈灵敏度图。通过将每个通道的线圈图像除以上述RSS图像，由此从重建好的校准帧图像计算出每个通道原始的灵敏度分布图。以这种方式计算的灵敏度图与机器扫描的图像几何参数相同，因此不需要配准。因此采用阈值处理从RSS图像中识别支持区域(阈值设置在0.1附近)，采用形态学成像方法填充灵敏度图中的空洞并平滑区域边界。本实施例用一个带有三次加权核的局部加权多项式回归(LWPR)拟合对灵敏度图中的支撑区域进行平滑去噪，对灵敏度图中的非支撑区域进行外推，得到优化的灵敏度图。本实施例采用的多项式为二阶多项式，支持区域的窗宽为12，非支持区域的窗宽为24。
3.再次，计算用来伪影抑制的校正因子图。利用优化的灵敏度图，在重建好的校准帧图像上进行加速因子为1的SENSE重建，生成参考图像ρ 0。然后在校准帧进行回顾性的欠采样，且回顾性欠采样的加速因子与后续需要重建的第二欠采样图像帧相同，再对欠采样后的数据进行SENSE重建，生成具有潜在混叠伪影的回顾性重建图像ρ 1。校正因子图定义为混叠图像逐点除以回顾性重建图像，即校正因子图中每个位置的像素值为参考图像和回顾性重建图像中对应位置像素值的商。此外，本实施例中的校正因子图需要被一个窗口为3×3的中值滤波 器进行滤波，以去除异常值。校正因子图C计算公式可以表述如下：
C＝ρ 0./ρ 1
4.最后，重建其余所有第二欠采样的图像帧。本步骤使用截断奇异值正则化方法对第二欠采样图像帧进行SENSE重建，丢弃小于最大值2％的奇异值。然后利用伪影抑制方法，将SENSE方法生成的图像与校正因子图逐点相乘，进一步减小误差，即可得到抑制伪影的完整图像帧。
需注意的是，由于本实施例中2D实验第二欠采样图像帧具有两种加速因子，即AF＝2和AF＝4，因此其在第3步中也需要分别基于AF＝2和AF＝4进行回顾性欠采样，生成两种不同的校正因子图，对应的校正因子图对应用于第4步中的第二欠采样图像帧SENSE重建。
为了进行比较，对回顾性欠采样的数据进行变加速的GRAPPA重建。在这种情况下，从第一欠采样图像帧的ACS得到的GRAPPA权重应用于所有其他帧。通过比较GRAPPA、KIPI与真实值的归一化均方根误差(RNMSE)来评估KIPI方法的准确性。在比较2D图像时，将完全采样图像作为真实值；对于3D图像时，由于全采样时间过长，将常规的2×1 GRAPPA图像视为真实值。
APTw参数图像计算如下。首先，将源图像配准到第一欠采样帧3.5ppm。其次，将两种不同TE获取的GRE图像的相位差计算为B0地图。第三，根据计算出的B0地图，为每个体素生成校正后的+3.5-ppm和-3.5-ppm信号值。最后，将校正后的3.5-ppm和+3.5-ppm图像相减得到APTw参数图像。
3、结果分析
从图1可以看出，GRAPPA方法(图1b,e)比KIPI方法(图1c,f)的误差要大得多(箭头)。KIPI生成高质量的图像(重建误差RNMSE＝0.008)。此外，在变加速因子欠采样的情况下，常规的GRAPPA在圈出地方的z谱(图1d中的虚线)与真实值(图1d中的实线)相比，产生了很大的误差。而KIPI产生的结果(图1d中的双划线)与真实值几乎一致。
图2展示了从源图像(图1)计算出的APTw参数图像。KIPI方法生成的APTw图像(图2c)与真实值非常一致(完全采样，图2a)，仅显示出轻微的差别。而使用AF＝4GRAPPA时，结果出现了大面积的低信号，带有明显的相位编码方向的混叠特征。KIPI和GRAPPA使用相同的变加速因子欠采样数据，选取第一 个+3.5ppm作为AF＝2的第一欠采样图像帧，选取S0和±3.5ppm(不包括第一欠采样帧)作为AF＝2，以及其他46帧AF＝4。
图3显示了在相同的实验条件下健康人类大脑的图像。与体模研究相似，结果也验证了KIPI的重建比GRAPPA更准确。一方面，KIPI生成的源图像(图3c)与真实值(图3a)的一致性优于使用GRAPPA从相同数据重建的源图像(图3b；重建误差分别为0.018和0.029)。另一方面，KIPI方法产生的z谱(图3d；双划线)与全k空间谱几乎无法区分(图3d；实线)，而GRAPPA导致显著误差(图3d；虚线)。
图4显示了从源图像(图3)生成的APTw参数图像。由于z谱的不准确，GRAPPA(图4b)的结果在图3(a)实线圈出对应的位置出现了明显的伪影。然而，与真实值(图4a)相比，KIPI方法(图4c)的图像质量几乎没有损失。这里看到的APTw图像缺乏增强是健康受试者的特征。KIPI和GRAPPA使用相同的变加速因子欠采样数据，选取第一个+3.5ppm作为AF＝2的第一欠采样图像帧，选取S0和±3.5ppm(不包括第一欠采样图像)作为AF＝2，以及其他46帧AF＝4。
图5显示了使用AF＝2×1(相位编码和层选编码方向)的传统GRAPPA获得的-4ppm的源图像，以及KIPI和常规GRAPPA重建的相同的欠采样数据的结果。由于扫描时间的限制，无法获得完全采样的3D CEST采集，因此传统的2×1 GRAPPA扫描(图5a)被认为是准确的真实值。尽管使用相同的变加速因子欠采样数据，KIPI(图5c)生成的源图像相比于GRAPPA(图5b)重建，更符合真实值。此外，在GRAPPA图像中，混叠伪影也很明显(图5b,d；箭头)。
图6为来自图5的源图像经过B0校正和图像配准后生成的APTw参数图像。在常规GRAPPA方法重建的APTw图像(图6b)中可以看到大量伪影，主要表现为层选方向上的混叠伪影。相比之下,KIPI方法生成的APTw图像(图6c)与真实值(图6a)几乎没有区别。变加速因子欠采样数据：+3.5ppm帧有ACS数据，AF＝2×2；其他6帧没有ACS数据且AF＝2×4。
图7显示了采用更高加速因子时的健康志愿者APTw参数图像。选取+3.5ppm帧作为第一欠采样图像帧，AF＝2×2，其他6帧的AF＝4×3。与图6(b)不同，这里GRAPPA的折叠伪影主要体现在相位编码方向上。同样，用GRAPPA重建的APTw图像存在明显的伪影，而在KIPI的结果上基本没有。KIPI单个源图像帧高达12的加速因子。总的来说，净有效加速因子可以达到8。
以上所述的实施例只是本发明的一种较佳的方案，然其并非用以限制本发明。有关技术领域的普通技术人员，在不脱离本发明的精神和范围的情况下，还可以做出各种变化和变型。因此凡采取等同替换或等效变换的方式所获得的技术方案，均落在本发明的保护范围内。

Claims (10)

  1. 一种结合k空间和图像空间重建的成像方法，用于多帧成像中欠采样图像帧进行重建，其特征在于，所述欠采样图像帧包含带有自动校准信号的第一欠采样图像帧和没有自动校准信号且加速因子不低于2的第二欠采样图像帧，且第一欠采样图像帧的加速因子不高于第二欠采样图像帧的加速因子；
    所述重建方法步骤如下：
    S1：针对第一欠采样图像帧，利用k空间中自校准的并行成像方法重建得到完整图像帧作为校准帧；对校准帧的k空间数据进行傅里叶变换后得到每个通道的线圈图像，合并所有通道的线圈图像后得到通道合并图像；
    S2：将每个通道的线圈图像除以所述通道合并图像，得到每个通道的线圈灵敏度图；从所述通道合并图像中识别出支撑区域，并对灵敏度图中的支撑区域进行平滑去噪，对灵敏度图中的非支撑区域进行外推，得到优化的灵敏度图；
    S3：利用优化的灵敏度图，在所述校准帧上进行加速因子为1的SENSE重建，生成参考图像；同时，对所述校准帧进行回顾性欠采样且回顾性欠采样的加速因子与所述第二欠采样图像帧的加速因子相同，再对欠采样后的数据进行SENSE重建，生成具有潜在混叠伪影的回顾性重建图像；基于参考图像和回顾性重建图像得到校正因子图，且校正因子图中每个位置的像素值为参考图像和回顾性重建图像中对应位置像素值的商；
    S4：针对第一欠采样图像帧之外的其余第二欠采样图像帧，利用所述优化的灵敏度图进行SENSE重建，将SENSE重建后的图像与所述校正因子图相乘，得到抑制伪影的完整图像帧。
  2. 如权利要求1所述的结合k空间和图像空间重建的成像方法，其特征在于，所述S1中，所述k空间中自校准的并行成像方法为GRAPPA方法，通过计算出线圈权重并应用于欠采样区域，重建得到完整图像帧。
  3. 如权利要求2所述的结合k空间和图像空间重建的成像方法，其特征在于，所述GRAPPA方法为带有Tikhonov正则化的GRAPPA。
  4. 如权利要求1所述的结合k空间和图像空间重建的成像方法，其特征在于，所述第一欠采样图像帧和第二欠采样图像帧均为二维图像；所述第一欠采样 图像帧的加速因子优选为2，所述第二欠采样图像帧的加速因子优选为2到4。
  5. 如权利要求1所述的结合k空间和图像空间重建的成像方法，其特征在于，所述第一欠采样图像帧和第二欠采样图像帧均为三维图像；所述第一欠采样图像帧的加速因子优选为2×2，所述第二欠采样图像帧的加速因子优选为N×M，其中2≤N≤4，2≤M≤4，且总加速因子不超过12。
  6. 如权利要求1所述的结合k空间和图像空间重建的成像方法，其特征在于，所述S1中，通过平方根方法或者自适应方法合并所有的线圈图像。
  7. 如权利要求1所述的结合k空间和图像空间重建的成像方法，其特征在于，所述S2中，对灵敏度图中的支撑区域通过拟合方式进行平滑去噪。
  8. 如权利要求1所述的结合k空间和图像空间重建的成像方法，其特征在于，所述S3中，所述校正因子图需经过滤波器滤波，以去除异常值。
  9. 如权利要求1所述的结合k空间和图像空间重建的成像方法，其特征在于，所述S4中，利用所述优化的灵敏度图进行SENSE重建时，采用截断奇异值正则化方法。
  10. 一种磁共振成像设备，其特征在于，包括磁共振扫描器以及控制单元，所述控制单元中存储有计算机程序，当所述计算机程序被执行时，用于实现如权利要求1～9任一项所述的成像方法；所述磁共振扫描器用于获取所述第一欠采样图像帧和第二欠采样图像帧数据。
PCT/CN2021/140280 2021-04-08 2021-12-22 结合k空间和图像空间重建的成像方法和装置 WO2022213666A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
JP2023558119A JP2024512529A (ja) 2021-04-08 2021-12-22 ｋ空間と画像空間とを組み合わせて再構築するイメージング方法及び装置
US18/482,883 US20240036141A1 (en) 2021-04-08 2023-10-08 Joint k-space and image-space reconstruction imaging method and device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110379060.9A CN113009398B (zh) 2021-04-08 2021-04-08 结合k空间和图像空间重建的成像方法和装置
CN202110379060.9 2021-04-08

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US18/482,883 Continuation US20240036141A1 (en) 2021-04-08 2023-10-08 Joint k-space and image-space reconstruction imaging method and device

Publications (1)

Publication Number Publication Date
WO2022213666A1 true WO2022213666A1 (zh) 2022-10-13

Family

ID=76388111

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/140280 WO2022213666A1 (zh) 2021-04-08 2021-12-22 结合k空间和图像空间重建的成像方法和装置

Country Status (4)

Country Link
US (1) US20240036141A1 (zh)
JP (1) JP2024512529A (zh)
CN (1) CN113009398B (zh)
WO (1) WO2022213666A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113009398B (zh) * 2021-04-08 2021-12-17 浙江大学 结合k空间和图像空间重建的成像方法和装置
CN113866694B (zh) * 2021-09-26 2022-12-09 上海交通大学 一种快速三维磁共振t1定量成像方法、系统及介质

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090001984A1 (en) * 2007-06-12 2009-01-01 Ken-Pin Hwang Method and apparatus for k-space and hybrid-space based image reconstruction for parallel imaging and artifact correction
DE102009014461A1 (de) * 2009-03-23 2010-09-30 Siemens Aktiengesellschaft Verfahren, Magnetresonanzgerät und Computerprogramm zur Erstellung von Bildern mittels paralleler Akquistionstechnik
CN104635188A (zh) * 2013-11-12 2015-05-20 上海联影医疗科技有限公司 K空间重建方法及磁共振成像方法
CN106491131A (zh) * 2016-12-30 2017-03-15 深圳先进技术研究院 一种磁共振的动态成像方法和装置
US20180081015A1 (en) * 2016-09-19 2018-03-22 Siemens Healthcare Gmbh Method and magnetic resonance apparatus for avoidance of artifacts in the acquisition of magnetic resonance measurement data
CN109839607A (zh) * 2019-01-10 2019-06-04 浙江大学 一种基于变加速敏感度编码的cest图像重建方法和装置
CN111513716A (zh) * 2019-02-05 2020-08-11 通用电气精准医疗有限责任公司 使用扩展灵敏度模型和深度神经网络进行磁共振图像重建的方法和系统
CN113009398A (zh) * 2021-04-08 2021-06-22 浙江大学 结合k空间和图像空间重建的成像方法和装置

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7482806B2 (en) * 2006-12-05 2009-01-27 Siemens Aktiengesellschaft Multi-coil magnetic resonance data acquisition and image reconstruction method and apparatus using blade-like k-space sampling
US7777487B2 (en) * 2007-02-15 2010-08-17 Uwm Research Foundation, Inc. Methods and apparatus for joint image reconstruction and coil sensitivity estimation in parallel MRI
IN2013CN00309A (zh) * 2010-07-02 2015-07-03 Koninkl Philips Electronics Nv
EP2656094A2 (en) * 2010-12-22 2013-10-30 Koninklijke Philips N.V. Rapid parallel reconstruction for arbitrary k-space trajectories
US9018952B2 (en) * 2011-05-27 2015-04-28 Mayo Foundation For Medical Education And Research Method for self-calibrated parallel magnetic resonance image reconstruction
US10089722B2 (en) * 2016-12-30 2018-10-02 Toshiba Medical Systems Corporation Apparatus and method for reducing artifacts in MRI images
US11143730B2 (en) * 2019-04-05 2021-10-12 University Of Cincinnati System and method for parallel magnetic resonance imaging

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090001984A1 (en) * 2007-06-12 2009-01-01 Ken-Pin Hwang Method and apparatus for k-space and hybrid-space based image reconstruction for parallel imaging and artifact correction
DE102009014461A1 (de) * 2009-03-23 2010-09-30 Siemens Aktiengesellschaft Verfahren, Magnetresonanzgerät und Computerprogramm zur Erstellung von Bildern mittels paralleler Akquistionstechnik
CN104635188A (zh) * 2013-11-12 2015-05-20 上海联影医疗科技有限公司 K空间重建方法及磁共振成像方法
US20180081015A1 (en) * 2016-09-19 2018-03-22 Siemens Healthcare Gmbh Method and magnetic resonance apparatus for avoidance of artifacts in the acquisition of magnetic resonance measurement data
CN106491131A (zh) * 2016-12-30 2017-03-15 深圳先进技术研究院 一种磁共振的动态成像方法和装置
CN109839607A (zh) * 2019-01-10 2019-06-04 浙江大学 一种基于变加速敏感度编码的cest图像重建方法和装置
CN111513716A (zh) * 2019-02-05 2020-08-11 通用电气精准医疗有限责任公司 使用扩展灵敏度模型和深度神经网络进行磁共振图像重建的方法和系统
CN113009398A (zh) * 2021-04-08 2021-06-22 浙江大学 结合k空间和图像空间重建的成像方法和装置

Also Published As

Publication number Publication date
CN113009398A (zh) 2021-06-22
US20240036141A1 (en) 2024-02-01
JP2024512529A (ja) 2024-03-19
CN113009398B (zh) 2021-12-17

Similar Documents

Publication Publication Date Title
Wang et al. Model‐based T 1 mapping with sparsity constraints using single‐shot inversion‐recovery radial FLASH
Jaubert et al. Water–fat Dixon cardiac magnetic resonance fingerprinting
Yang et al. Sparse reconstruction techniques in magnetic resonance imaging: methods, applications, and challenges to clinical adoption
US11143730B2 (en) System and method for parallel magnetic resonance imaging
Cheng et al. Nonrigid motion correction in 3D using autofocusing withlocalized linear translations
Xue et al. High spatial and temporal resolution retrospective cine cardiovascular magnetic resonance from shortened free breathing real-time acquisitions
US9886745B2 (en) Multi-shot scan protocols for high-resolution MRI incorporating multiplexed sensitivity-encoding (MUSE)
US9797974B2 (en) Nonrigid motion correction in 3D using autofocusing with localized linear translations
Fair et al. A review of 3D first-pass, whole-heart, myocardial perfusion cardiovascular magnetic resonance
Küstner et al. MR image reconstruction using a combination of compressed sensing and partial Fourier acquisition: ESPReSSo
US10739432B2 (en) Dynamic magnetic resonance imaging
JP2008539852A (ja) 磁気共鳴イメージングシステムのそれぞれの信号チャネルにおける独立した運動補正
Haris et al. Free‐breathing fetal cardiac MRI with doppler ultrasound gating, compressed sensing, and motion compensation
US20240036141A1 (en) Joint k-space and image-space reconstruction imaging method and device
US11002815B2 (en) System and method for reducing artifacts in echo planar magnetic resonance imaging
US20240095889A1 (en) Systems and methods for magnetic resonance image reconstruction with denoising
Bonanno et al. Self-navigation with compressed sensing for 2D translational motion correction in free-breathing coronary MRI: a feasibility study
Liu et al. Single multi-echo GRE acquisition with short and long echo spacing for simultaneous quantitative mapping of fat fraction, B0 inhomogeneity, and susceptibility
US10852381B2 (en) Susceptibility mapping of a moving object
Merrem et al. Rapid diffusion-weighted magnetic resonance imaging of the brain without susceptibility artifacts: Single-shot STEAM with radial undersampling and iterative reconstruction
Gallo-Bernal et al. Pediatric magnetic resonance imaging: faster is better
Constantinidesa et al. Restoration of low resolution metabolic images with a priori anatomic information: 23Na MRI in myocardial infarction☆
US20160054420A1 (en) Compensated magnetic resonance imaging system and method for improved magnetic resonance imaging and diffusion imaging
Wang Arterial spin labeling perfusion MRI signal processing through traditional methods and machine learning
Liu et al. 3D true-phase polarity recovery with independent phase estimation using three-tier stacks based region growing (3D-TRIPS)

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21935882

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 2023558119

Country of ref document: JP

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21935882

Country of ref document: EP

Kind code of ref document: A1