Travel

A correction method for streak artifacts in gradient-echo EPI using spin-echo EPI reference data

Description
To analyze the streak artifacts in a gradient-echo echo planar imaging (GE-EPI) sequence and to propose a correction method for the Nyquist ghost artifacts that does not cause streak artifacts in the GE-EPI imaging. Several GE-EPI imaging experiments
Categories
Published
of 9
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  RESEARCH ARTICLE A correction method for streak artifacts in gradient-echo EPIusing spin-echo EPI reference data Jun-Young Chung  • Yeji Han  • Zang-Hee Cho  • HyunWook Park Received: 24 May 2011/Revised: 5 October 2011/Accepted: 10 October 2011/Published online: 10 November 2011   ESMRMB 2011 Abstract Objective  To analyze the streak artifacts in a gradient-echo echo planar imaging (GE-EPI) sequence and to pro-pose a correction method for the Nyquist ghost artifactsthat does not cause streak artifacts in the GE-EPI imaging.  Materials and methods  Several GE-EPI imaging experi-ments with various reference scans, using both GE-EPI andSE-EPI scan data, were performed to analyze the streak artifacts and to investigate the spin dephasing phenomenaof the GE-EPI reference scan. In addition, the analysisbased on the spin dephasing was undertaken in order todemonstrate that the SE-EPI reference data can be used forthe correction of the GE-EPI main scan data.  Results  The experimental results confirmed that theimprovement of the reference data using either signalaveraging or a large flip angle cannot guarantee perfectcorrection of the streak artifact if the noise is not com-pletely removed. Due to the main field inhomogeneity, thespins of the GE-EPI reference data were dephased inmultiple echo signals. The proposed correction method,which uses a SE-EPI reference scan for the GE-EPI ima-ges, eliminates the N/2 ghost artifacts without producingstreak artifacts. Conclusion  It is believed that the proposed phaseerror correction scheme can improve the EPI performancein high field MRIs with higher magnetic fieldinhomogeneities. Keywords  Echo planar imaging (EPI)    Streak artifacts   Nyquist ghost artifact    Gradient-echo EPI    Spin-echo EPIreference scan    Field inhomogeneity in high field MRI Introduction Echo planar imaging (EPI) [1] is a widely used magneticresonance imaging (MRI) technique that is especiallysuitable for applications requiring a short acquisition time,such as functional magnetic resonance imaging (fMRI),diffusion-weighted imaging (DWI), and dynamic imaging.Although EPI is favored for its time efficiency, the lim-ited spatial resolution, Nyquist ghost artifacts, and geo-metric distortions of the EPI have become a majorconcern in many application areas. The Nyquist ghostartifact of EPIs is inherent as the even and odd phaseencoding lines of the k-space are successively sampled inopposite readout directions in the presence of variousproblems such as field inhomogeneity, hardware imper-fections, and eddy currents [2–4]. The geometric distor- tions are also caused by field inhomogeneities, includingthe large static field inhomogeneity and the local mag-netic field changes that result from various factors such assusceptibility differences at the tissue/air and tissue/boneinterfaces [5–10]. Since a larger susceptibility difference results in a larger variation of the local magnetic field,these artifacts become more critical in 3 T and higherfield MR systems.In order to correct the above-mentioned EPI artifacts, anumber of different methods have been suggested. Forexample, the geometric distortions can be corrected using J.-Y. Chung    Z.-H. ChoNeuroscience Research Institute,Gachon University of Medicine and Science,Incheon, Republic of KoreaY. Han    H. Park ( & )Department of Electrical Engineering, Korea Advanced Instituteof Science and Technology, 373-1 Guseong-dong, Yuseong-gu,Daejeon 305-701, Republic of Koreae-mail: hwpark@kaist.ac.kr  1 3 Magn Reson Mater Phy (2012) 25:205–213DOI 10.1007/s10334-011-0289-0  an estimated value of the B 0  field inhomogeneity with agradient echo (GE), an offset spin echo (SE), or EPI imageswith different echo times (TE) [11–13]. For the Nyquist ghost artifact corrections, phase correction can be per-formed using the full reference data acquired with orwithout phase encodings [14–19] or using only two lines of  a reference scan [20]. For example, Bruder et al. [14] used a reference scan without the phase encoding gradient andthe correction was performed for all echoes using the phaseinformation obtained from the reference scan. Hu andTuong [19] used two sets of imaging scans acquired usingreversed readout gradients, and the k-space lines withpositive and negative readout gradients were rearrangedfrom two different scans. Heid [20] used either one even orone odd line from a reference scan for correction. Othertechniques have performed corrections based on post-pro-cessing images [21–25]. Instead of acquiring an additional reference scan, Buonocore and Gao [21] reconstructed twoimages from one set of EPI data: one from the even-echosignals with zero filling on the odd-echo positions and theother from the odd-echo signals with zero filling on theeven-echo positions. From the two reconstructed images,the phase information was extracted and used to compen-sate the phase errors. Lastly, there have also been 2Dcorrection approaches that exploit 2D information fromadditional full k-space reference scans [27–31]. The above-mentioned phase correction techniques canbe categorized into linear or nonlinear methods. Linearcorrection methods typically perform linear fitting to theacquired phase information and only the spatially constantand linear (first order) phase errors are compensated.However, the nonlinear correction methods perform point-by-point phase corrections along the readout direction(  x  direction) in the (  x ,  k   y ) space. Thus, it is not constrainedto perform corrections using only the constant and linear(first order) spatial variations in the phase errors. As aresult, linear correction methods, e.g. Heid’s method [20]or Buonocore’s method [21], are less accurate and limitedin their ability to correct ghost artifacts compared withnonlinear methods using a full reference scan [14, 19]. Consequently, nonlinear methods are preferred in manycases, especially in high field MRIs. However, a sharpchange in the phase due to phase error, i.e. an abrupt dis-crepancy in the neighboring data along the readout direc-tion after the nonlinear correction, may generate streak artifacts in the reconstructed image [19, 26]. Hu and Tuong anticipated that the streak artifacts were most likely theresult of severe field inhomogeneity rather than the resultof vascular or cerebrospinal fluid (CSF) flows as the streakswere also present in the phantom data. However, the exactcause has not yet been clearly explained.Since a solution for the streak artifacts in nonlinearphase correction has not yet been developed, the streak artifacts are analyzed in this study in order to locate anaccurate source and a new method is proposed to eliminatethe streak artifacts using SE-EPI reference scans. Materials and methods Data acquisitionThe images were acquired from a spherical water phantom(containing NiSO 4 ) and a human brain with 1.5 Tesla(Avanto, Siemens Medical Solutions, Erlangen, Germany)and three Tesla MRI scanners (ISOL Technology, Korea)using a quadrature head coil. For the analysis of the streak artifacts, the reference and main data were both acquiredusing single-shot GE-EPI and single-shot SE-EPI sequen-ces. With the 1.5 Tesla system, the scans were performedusing the following parameters: TE  =  40 ms for GE-EPIand SE-EPI, TR  =  3,000 ms, matrix size  =  64  9  64,number of slices  =  36, slice thickness  =  5 mm, FOV  = 256  9  256 mm 2 , number of averages  =  1, bandwidthper pixel  =  2,605 Hz/Px, ramp sampling  =  on, gradientshape  =  trapezoidal, and phase encoding direction  = anterior/posterior (A/P). In order to compensate for thefield homogeneity, 3D nonlinear shimming was performedusing a built-in procedure. For the 3 Tesla system,the following parameters were used: TE  =  35 ms, TR  = 3,000 ms, matrix size  =  64  9  64, number of slices  =  15,slice thickness  =  4 mm, FOV  =  256  9  256 mm 2 , numberof averages  =  1, bandwidth per pixel  =  1,953 Hz/Px,ramp sampling  =  on, gradient shape  =  trapezoidal, andphase encoding direction  =  A/P. The imaging volume onthe 3 T system was shimmed to a full width at a half maximum of approximately 20 Hz.The phase correction procedures were then performedusing different reference data in order to investigate theeffectofthephaseoffset.Inthisstudy,thephaseinformationof the projection data was measured for the reference data.That is, the phase of the data along the  x  direction after the1D Fourier transform (FT) of the echo signals was measuredand the measured phase term was subtracted point-by-pointfrom the projection data. For each dataset, four differentimages were reconstructed using different combinations of GE-EPI and SE-EPI sequences, i.e. GE main scan/GE ref-erence scan, SEmain scan/SE referencescan, SEmain scan/ GE reference scan, and GE main scan/SE reference scan.Analysis of the streak artifactsIn this study, a point-by-point phase correction techniquewas applied to the acquired data after the 1D Fouriertransform (FT), i.e. the correction was performed along thereadout direction (  x  direction) in the (  x ,  k   y ) space. If there 206 Magn Reson Mater Phy (2012) 25:205–213  1 3  was a phase error at a certain point in the (  x ,  k   y ) space afterthe nonlinear phase correction, a streak artifact was gen-erated along the corresponding  x  position in the recon-structed image. In order to verify the exact cause of thephase error that results in the streak artifact, the focus wasplaced on the spin dephasing that occurs during the readoutperiod or during the echo series of both the main and ref-erence data collection in the GE-EPI sequence. Due to themain field inhomogeneity, the spins of the GE-EPI refer-ence data inevitably experienced dephasing in multipleecho signals, and the magnitude of the projection datacould be influenced such that some data become zero ornear-zero. These zero or near-zero magnitudes can result ina significant error when used for phase estimation. Whenthe reference data is acquired with zero phase encodinggradients, the reference data after the 1D-FT (in the readoutdirection) corresponds to the projection data along thephase encoding direction. In order to demonstrate the effectof the spin dephasing due to the field inhomogeneity, twosets of images were acquired with different shim valuesand the magnitudes of the different GE-EPI reference dataafter the 1D-FT were calculated as shown in Fig. 1a, b. Asthe datasets were obtained with different magnetic fieldshim values, the positions of the zero magnitude in theprojection data were changed as marked with arrows in thefigures, and thus different streak artifacts were generated atthe corresponding locations as illustrated in Fig. 1c, d. As aresult, it can be concluded that the streak artifacts aregenerated due to the (near-) zero magnitude of the pro- jection data caused by the spin dephasing.Phase offset between GE-EPI and SE-EPIAssuming the same field inhomogeneity,  ~ S   Rn  k  RO ð Þ  and ~ S   M n  k  RO ð Þ  represent the k-space signals of the reference andmain EPI scans, respectively, where the subscript  n  denotesthe phase encoding or echo number. Similarly, the corre-sponding ideal k-space signals, i.e. without distortionscaused by the field inhomogeneity, are denoted as  S   Rn  k  RO ð Þ and  S   M n  k  RO ð Þ , respectively. When the 1D-FT was per-formed along the readout direction (  x  direction), the cor-responding 1D transform of the above-mentioned signalscan be denoted as  P  Rn  x ð Þ ,  ~ P  Rn  x ð Þ ,  P  M n  x ð Þ , and  ~ P  M n  x ð Þ . Byassuming a phase error  U  x ð Þ  for the reference and mainsignals, the following equations can be defined: =  1 ~ S   Rn  k  RO ð Þ    ¼  ~ P  Rn  x ð Þ ¼  P  Rn  x ð Þ e i U  Rn  x ð Þ =  1 ~ S   M n  k  RO ð Þ    ¼  ~ P  M n  x ð Þ ¼  P  M n  x ð Þ e i U  M n  x ð Þ ;  ð 1 Þ where  U  Rn  x ð Þ  and  U  M n  x ð Þ  are the corresponding phasecomponents of the  ~ P  Rn  x ð Þ  and  ~ P  M n  x ð Þ , respectively, and  =  1 is the inverse Fourier transform operator.Since the ideal projection data has positive real values,the nonlinear phase term  U  Rn  x ð Þ  can be deduced from e i U  Rn  x ð Þ ¼  ~ P  Rn  x ð Þ ~ P  Rn  x ð Þ j j . Using the phase term derived from thereference scan, one can derive the final phase correctedsignal as follows,  ^ P  M n  x ð Þ ^ P  M n  x ð Þ ¼  ~ P  M n  x ð Þ ~ P  Rn  x ð Þ  ~ P  Rn  x ð Þ¼  P  M n  x ð Þ e i U  M n  x ð Þ e  i U  Rn  x ð Þ ¼  P  M n  x ð Þ e i  U  M n  x ð Þ U  Rn  x ð Þ ð Þ :  ð 2 Þ Equation 2 suggests that the signal after the phasecorrection would become ideal projection data,  P  M n  x ð Þ , if the  e i  U  M n  x ð Þ U  Rn  x ð Þ ð Þ term can be removed. If the referenceand main scans are acquired using the same sequence,except the phase encoding gradients under the samemagnetic field conditions (under the same shimconditions),  U  Rn  x ð Þ  U  M n  x ð Þ  becomes zero and theestimated projection data,  ^ P  M n  x ð Þ , is equal to the idealprojection data,  P  M n  x ð Þ : This property can also be applied when different typesof sequences, i.e. GE or SE, are used separately for themain and reference scans if the nonlinear phase terms of the main and reference scans have a constant phase offset.For example, if the main scan data acquired with the GE-EPI has a constant phase offset,  / 0 , from the referencescan data acquired with the SE-EPI (see Fig. 2), it can bedescribed as  U  M n  x ð Þ  U  Rn  x ð Þ ¼  / 0 . Thus, the SE-EPIsequence can also be used as a reference scan for thephase correction of the GE-EPI data. For the SE-EPIreference scan, the final estimated signal can be written asfollows: ^ P  M n  x ð Þ ¼  P  M n  x ð Þ e i  U  M n  x ð Þ U  Rn  x ð Þ ð Þ¼  P  M n  x ð Þ e i / 0 :  ð 3 Þ Then, the constant phase term,  / 0 , in Eq. 3 can beremoved using the magnitude in the image space after theFT along the phase encoding direction.The phase offset,  / 0 , between the GE-EPI and SE-EPI,is illustrated in Fig. 2 using a phase evolution diagram.These two diagrams demonstrate how phase errors areaccumulated as time evolves in the spin phase of the GE-EPI and SE-EPI. As shown in Figs. 2a and b, the phaseerror in the GE-EPI is identical to that of the SE-EPI,except for the constant phase offset, / 0 . Thus, both the GE-EPI and SE-EPI reference data can be used to correct forthe GE-EPI or the SE-EPI main scans. Magn Reson Mater Phy (2012) 25:205–213 207  1 3  Results Reference data analysisFrom the main and reference scan data acquired from thefour different combinations of GE-EPI and SE-EPIsequences at 1.5 T, four images were reconstructed asshown in Fig. 3. Although the ghost artifacts are mostlyremoved in all cases, the two images corrected using theGE-EPI reference data contain streak artifacts (Fig. 3a, c).However, the streak artifacts were completely eliminated(see Fig. 3b, d) in the two other images that were correctedusing the SE-EPI reference data.Ideally, if the magnetic field is perfectly homogeneousin the imaging region, an identical set of reference datashould be acquired from the GE-EPI and SE-EPI scans.In the actual experiments, however, the data acquired bythe GE-EPI and SE-EPI reference scans were different:for the GE-EPI reference sequence, the echoes progres-sively broadened (spin disperse) and decreased in size asthe readout time increased. For the SE-EPI referencesequence, the echoes were more focused at the center of the k-space compared with the GE-EPI due to the refo-cusing property of the 180   RF pulse. As a result, theimages corrected using the SE-EPI reference scan clearlydemonstrated a major improvement as the streak artifactswere completely eliminated as shown in Fig. 3b, d. Inorder to verify the mathematical analysis presented in thematerials and methods section, the projection data of theGE-EPI and SE-EPI reference data are represented asmagnitudes (Fig. 3e, f) and as profiles of the three dif-ferent positions (Fig. 3g, h). Due to the spin dephasing,the profiles of the GE-EPI have several near-zero values,which could cause the streak artifacts, as marked witharrows in Fig. 3e, g. However, the profiles of the SE-EPIdo not have zero values in the projection data. Fromthese experiments, it was confirmed that the phase cor-rection with ‘‘near-zero value’’ data can result in streak artifacts. Fig. 1 a ,  b  Magnitudes of theGE-EPI reference data (theprojection data) arranged insequence (time series) after 1D-FT of the echo signals acquiredat two different shimmingvalues.  c ,  d  The correctedimages from the reference data( a ) and ( b ) using a nonlinearphase correction technique Fig. 2  Spin phase diagrams of two different EPI sequences.  a  GE-EPI and  b  SE-EPI208 Magn Reson Mater Phy (2012) 25:205–213  1 3  Phase correction resultsFigure 4 shows the original and the corrected phantomimages of the EPI data acquired in the 1.5 T (left) and 3 T(right) MRI systems. The uncorrected original imagecontains the N/2 ghost artifacts as shown along the phaseencoding direction in Fig. 4a. When the GE-EPI referencedata was used for the phase correction, most N/2 ghostartifacts were removed while also generating a new streak artifact due to the phase correction error (Fig. 4c). Fig. 3  Corrected images fromfour combinations of thereference and main scans usinga 1.5 Tesla system.  a  Acorrected image of the GE-EPImain scan using the GE-EPIreference data.  b  A correctedimage of the SE-EPI main scanusing the SE-EPI reference data. c  A corrected image of the SE-EPI main scan using the GE-EPIreference data.  d  A correctedimage of the GE-EPI main scanusing the SE-EPI reference data. e  Magnitude (projection data) of the GE-EPI reference data after1D-FT.  f   Magnitude of the SE-EPI reference data after 1D-FT. d ,  g  Three selected profilesof ( e ).  h  Three selectedprofiles of ( f  )Magn Reson Mater Phy (2012) 25:205–213 209  1 3
Search
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks