RASER: A New Ultrafast Magnetic Resonance ImagingMethod

Ryan Chamberlain,1 Jang-Yeon Park,1 Curt Corum,1 Essa Yacoub,1 Kamil Ugurbil,1

Clifford R. Jack Jr,2 and Michael Garwood1*

A new MRI method is described to acquire a T2-weighted imagefrom a single slice in a single shot. The technique is based onrapid acquisition by sequential excitation and refocusing(RASER). RASER avoids relaxation-related blurring because themagnetization is sequentially refocused in a manner that effec-tively creates a series of spin echoes with a constant echo time.RASER uses the quadratic phase produced by a frequency-swept chirp pulse to time-encode one dimension of the image.In another implementation the pulse can be used to excitemultiple slices with phase-encoding and frequency-encoding inthe other two dimensions. The RASER imaging sequence ispresented along with single-shot and multislice images, and iscompared to conventional spin-echo and echo-planar imagingsequences. A theoretical and empirical analysis of the spatialresolution is presented, and factors in choosing the spatialresolution for different applications are discussed. RASER pro-duces high-quality single-shot images that are expected to beadvantageous for a wide range of applications. Magn ResonMed 58:794–799, 2007. © 2007 Wiley-Liss, Inc.

Key words: single-shot imaging; MRI; chirp; spatial encoding;time-encoding; quadratic phase profile; frequency-swept pulse

Traditional ultrafast MRI sequences collect multiple ech-oes per excitation. Although this approach dramaticallyreduces image acquisition time, it does suffer from signif-icant drawbacks. Image resolution can be degraded be-cause each echo is acquired at a different echo time (TE)and the transverse relaxation (T2 or T2*) decay of theechoes acts as an apodization filter in k-space (1). Echo-planar imaging (EPI), in particular, suffers from significantimage blurring due to the relatively short effective trans-verse relaxation time (T2*). Furthermore, images recon-structed from gradient echo trains often exhibit geometricdistortion and signal dropout in regions where steep mag-netic field gradients are present in the object (e.g., due tomagnetic susceptibility interfaces). Here, a new pulse se-quence is described in which multiple echoes per excita-tion are acquired in such a way that all echoes have thesame T2-weighting and thus image blurring from T2 or T2*decay does not occur. This method is called RASER sinceit is based on rapid acquisition by sequential excitation

and refocusing. By producing a train of locally refocusedspin echoes, RASER avoids many of the image artifactscommon to techniques based on rapid gradient reversal toform echoes, like EPI.

Other groups have described pulse sequences that, likeRASER, accomplish sequential excitation and/or refocus-ing of spins along the direction of an applied field gradi-ent. Meyerand and Wong (2) introduced a time-encodingsequence that uses a series of 90° pulses to excite differentslices, followed by a single 180° pulse to create spin ech-oes that are read out in opposite order. Although thissequence produces a series of spin echoes, the TE value ofthe different echoes varies widely across the object. Morerecently, Frydman and co-workers (3,4) introduced a MRImethod that exploits a chirp pulse (5) to spatially encodeone dimension, instead of frequency encoding. Althoughthis method makes use of a novel spatial-encodingscheme, the achievable image quality is also limited by T2*decay during the echo train. RASER, on the other hand,combines the ideas of these two sequences by using a chirpexcitation pulse to excite magnetization sequentially alongone spatial axis, followed by two 180° pulses to create T2

weighting. The signals produced are then read out sequen-tially in time and the image is reconstructed by performinga Fourier transformation (FT) in the frequency-encodeddirection only. RASER avoids susceptibility-induced im-age artifacts and blurring due to T2 (or T2*) decay becausethe echoes all have the same effective TE, and the resultingT2 contrast is constant throughout the image.

THEORY

A multislice version of the RASER pulse sequence isshown in Fig. 1a. In the present implementation, excita-tion is performed with a long duration 90° chirp pulse thatis applied in the presence of a field gradient (called thetime-encoding gradient, Gte). The excitation pulse pro-duces a 90° rotation for all magnetization in a slab havinga thickness �x equal to �2�bw/�Gte�, where bw is the band-width of the pulse (in Hz) and � is the gyromagnetic ratio.Figure 2 shows an example of the excitation profile pro-duced by a 90° chirp pulse using bw� 20 kHz and durationTp � 0.0275 sec. In chirp, the RF varies linearly in timeand, therefore, in the direction of Gte, isochromats undergo90° rotations at different times. Once excited, a given iso-chromat precesses in the transverse plane for the rest of thepulse (6,7). In this way, the transverse phase of each iso-chromat created by the presence of Gte evolves a differentamount. Following excitation, a pair of 180° pulses to-gether with additional Gte pulses are used to reverse thisphase evolution in a sequential manner during the acqui-sition period. In the multislice version of RASER (Fig. 1a),

1Center for Magnetic Resonance Research and Department of Radiology,University of Minnesota Medical School, Minneapolis, Minnesota.2Department of Radiology, Mayo Clinic College of Medicine, Rochester, Min-nesota.Grant sponsor: National Institutes of Health (NIH); Grant numbers: P41RR008079, P30 NS057091, R01 CA92004, R01 MH070800; Grant sponsor:MIND Institute.*Correspondence to: Michael Garwood, PhD, Center for Magnetic ResonanceResearch, 2021 Sixth St. SE, Minneapolis, MN 55455.E-mail: gar@cmrr.umn.eduReceived 31 October 2006; revised 12 June 2007; accepted 25 July 2007.DOI 10.1002/mrm.21396Published online in Wiley InterScience (www.interscience.wiley.com).

Magnetic Resonance in Medicine 58:794–799 (2007)

© 2007 Wiley-Liss, Inc. 794

a phase-encoding gradient along a second spatial axis isalso needed to generate complete image data.

After excitation by the chirp pulse, the phase of themagnetization (�) will have a quadratic dependence onposition (x) in the direction of Gte. At the vertex of thequadratic phase function, d�/dx � 0; therefore, only iso-chromats on opposite sides of a plane defined by the vertexwill have similar phase. Assuming that the distribution ofmagnetization is locally symmetric when reflected about agiven vertex, measurable signal will arise predominatelyfrom the isochromats close to the plane of the vertex. Thesteepness of the � profile, and therefore the localizingeffect of the profile, is determined by the parameter, R �bw � Tp (7).

Following excitation, the combination of gradients and180° pulses ensures that the quadratic phase profile ispreserved. In other words, when the acquisition periodbegins the vertex of the quadratic phase is at the edge ofthe slab that was excited first. Continued application of thetime-encoding gradient moves the vertex across the slab,so the signal originates from sequential planar locations asthe acquisition proceeds. After the acquisition period(Tacq) has lasted for a period equal to Tp, the vertex of thequadratic phase profile reaches the edge of the slab thatwas excited last. During Tacq, signal localization is accom-plished in the time-encoded direction simply by acquiringat sequential time points (i.e., no FT is performed).

In the acquisition period the vertex position is movedalong the time-encoded direction by either a series of Gte

blips between readouts (Fig. 1a) or by applying a constantGte (Fig. 1b,c). With blipped Gte the vertex is stationaryduring the acquisition of individual frequency-encodedlines and is shifted between these readouts. However, theversion with continuous Gte is less demanding of the gra-dient hardware, and it allows more echoes to be acquired

for a given amount of time. The drawback is that it requiresadditional postprocessing steps to remove the effect of Gte

on the frequency-encoded lines. Specifically, due to thepresence of Gte a time-dependent linear phase must beapplied to each data point in the frequency-encoded di-rection prior to FT. This phase shift is given by:

�� � –2��Gte�xnrosw�1nte/Nte–xc/�x–0.5 [1]

where �x is the slab thickness, xc is the center position ofthe slab excited by the chirp pulse, nro is an integer indexrepresenting a given point in the read-out direction, nte isan integer index representing the current point in thetime-encoded direction, Nte is the total number of points inthe time-encoded direction, and sw is the bandwidth of thefrequency-encoding.

The quadratic phase profile produced across a slab ex-cited by a chirp pulse was analytically described by Pipe(6). Following Pipe’s derivation, during acquisition thequadratic phase across the slab can be shown to be:

�� � �R–�2 � 2�–4��/4 [2]

where � is normalized position (�x–xc/�x) and � is thefractional area of the time-encoding gradient that has beenbalanced. In the version of RASER with constant Gte ac-quisition, � � t/Tacq, where t is the length of time that Gte

has been applied since the final 180° pulse. During Tacq,the presence of Gte changes �, thereby shifting the vertexcontinuously in time (7). By expressing Eq. [2] in terms ofnormalized position, it can be seen that the phase profiledepends only on the value of R.

RASER is effectively a double-echo sequence with aunique type of “sequential” spin echo created at eachvertex location. RASER differs from a conventional Hahnspin-echo sequence in that the isochromats are locallyrefocused sequentially in time, as opposed to being glo-bally refocused at one instant in time. In the present con-text, the local refocusing time is defined as the point intime when isochromats circa a given position x have beenrefocused (i.e., the time at which magnetic susceptibility,static field inhomogeneity, and chemical shifts have been

FIG. 1. a: Multislice RASER pulse sequence with a blipped gradientin the acquisition period (Tacq). b,c: Single-shot RASER sequenceusing a continuous time-encoding gradient in the acquisition time,Tacq. The TE is the same for the isochromats highlighted in (b) and(c). As shown, there are 14 echoes collected, resulting in 14 pixelsin the time-encoding direction. The number of echoes can bechanged to suit the experiment.

FIG. 2. Bloch simulation of the 90° excitation profile of a chirp pulse(R � 550) having a bandwidth (bw) of 20 kHz. This pulse produces arelatively flat 90° excitation profile for isochromats within the band-width of the frequency sweep. The amplitude of the chirp pulse(�B1/2�) was 230 Hz.

RASER: New Ultrafast MRI Method 795

locally compensated by the pair of 180° pulses). The localrefocusing time is analogous to the TE of a conventionalspin echo, and it follows the same timing rules. In RASER,Tacq begins at the local refocusing time of the isochromatthat was excited first and it ends at the local refocusingtime of the isochromat that was excited last. During Tacq,the local refocusing time occurs at sequential values of xthat coincide with sequential vertex locations, so the sig-nal always originates predominantly from locally refo-cused (in-phase) magnetization.

The single-shot version of RASER is similar to the mul-tislice version except that it has no phase-encoding gradi-ents, and it requires a slice-selection gradient during atleast one of the 180° pulses. In multislice RASER, thetime-encoded direction is through-plane, while in single-shot RASER the time-encoded direction is in-plane. Thesingle-shot RASER sequence is depicted for two differentisochromats in Fig. 1b,c.

RASER offers the capability for restricted field-of-view(FOV) in the time-encoding direction (i.e., to zoom), whileavoiding some of the image artifacts that often plague suchtechniques (e.g., aliased signals). Thus, in RASER the FOVin the time-encoded dimension is set by the slab thickness,so there is no need to set the FOV bigger than the slab toprevent folding near its borders.

Simulations

The point spread function (PSF) in the time-encoded di-rection was estimated using numerical simulations. Forthe simulation the relative transverse magnetization (Mxy)in a time-encoding of a 1D Gaussian-shaped object wascalculated based on Bloch equations. The 1D image of theobject was generated from the set of Mxy values sampledduring Tacq (Nte � 256). Figure 3a shows the object, as wellas time-encoded 1D images of the object, using twodifferent R values. As expected, a better representationof the object (i.e., due to higher resolution) is obtainedwith the larger R value. The PSF was calculated from thetime-encoded image using the convolution theorem (8):

P � S /H [3]

where P( ), S( ), and H( ) are FTs of the PSF, image, andobject, respectively, and is the frequency variable. Asshown in Fig. 3b, the numerically determined PSF is asmooth function. Furthermore, in agreement with the the-oretical description given above (Eq. [2]), the width of thePSF in terms of the normalized spatial coordinate � de-pends only on the value of R. To obtain an approximate Rdependence for the normalized PSF width, a least-squarefitting of the numerically determined PSF was performed,which yielded:

��PSF � 17.0/R � 0.00039 [4]

where ��PSF is the full-width at half-maximum (FWHM) ofthe normalized PSF. Hence, in multislice RASER a givenslice thickness can be obtained using:

R � 17.0/thk/�x–0.00039 [5]

where thk is the desired slice thickness (FWHM). Like-wise, the R value needed to achieve a given in-plane spa-tial resolution (Nte�1/��PSF) in a single-shot RASER exper-iment can be estimated using:

R � 17.5Nte–0.55 [6]

Thus, for 32 contiguous pixels R needs to be 550.Simulations based on the Bloch equations were also

used to estimate the relative signal energy contained in atime-encoded acquisition versus a 1D gradient-echo acqui-sition, without considering magnetic susceptibility or re-laxation effects. The simulated 1D object was composed ofM isochromats distributed evenly over a frequency rangeof �10 to 10 kHz in 10-Hz increments (M � 2001). The

FIG. 3. a: The 1D object (solid) and simulated time-encoded 1Dimages (dashed) for R � 550 and R � 400. b: The theoretical (solid)and experimental (dashed) PSF in the time-encoded direction for anexcitation pulse with R � 550. The FWHM of the theoretical PSF is1 pixel wide, and the FWHM of the Gaussian fit to the experimentalPSF is 1.33 pixels wide in a 32 pixel image. c: The FWHM of the PSFfor R � 400, 550, and 700 obtained from theoretical and experi-mental data. The FWHM of the theoretical PSF is calculated fromthe Gaussian fit. d: The 2D image of the PSF phantom used for theexperimental PSF measurements (R � 550). The Teflon block oc-cupies the center of the image. The flat surface used for the PSFmeasurement is shown by the black arrow.

796 Chamberlain et al.

evolution of the transverse magnetization was numericallycalculated using the proper gradient waveforms and datasampling rates for the two sequences. The number of sam-pled data points in both sequences (N) was 32, with sam-pling interval (dw) of 859 �s (� Tp/Nte) and 50 �s (�1/sw � 1/(20 kHz)) in the time-encoded and gradient-echoacquisitions, respectively. The time-encoded and gradient-echo sequences used a 90° chirp (R � 550, Tp � 27.5 ms,�B1/2� � 230 Hz) and a 90° delta pulse for excitation,respectively. The relative signal energy (Esignal) was esti-mated using:

Esignal � dw�j�1

N ���i�1

M

Mxi,j�2

� ��i�1

M

Myi,j�2� , [7]

where Mx(i,j) and My(i,j) are the transverse magnetizationcomponents of the ith isochromat at the time when the jth

data point is sampled. The Bloch simulations predictednearly identical Esignal in both types of acquisitions (within1%). In the time-encoded simulation the signal energy wasequally distributed throughout the data points. Although1D FT is necessary to form the 1D image from the gradient-echo data, theory predicts the signal energy to be un-changed after performing FT, in accordance with Parse-val’s theorem (8).

Bloch simulations were also used to estimate the relativespecific absorption rate (SAR) of sequences using the samerepetition time (TR). The average SAR of the followingsequences was estimated: RASER, conventional spin-echoEPI, multislice spin-echo, and fast spin-echo using 180°refocusing pulses. RASER calculations used the samechirp pulse described above, whereas the 90° pulse in theother spin-echo sequences was a 4 ms sinc pulse (�B1/2� �350 Hz). The 180° pulses in all calculations were 4 ms sincpulses (�B1/2� � 700 Hz). The average SAR in RASER wasa factor of 5.28 greater than that of spin-echo EPI and wasless than that in multislice spin echo once the number ofslices was greater than 5. The average SAR in RASER wasless than that in fast spin echo once the echo train lengthwas greater than 6.

MATERIALS AND METHODS

All experiments were performed with a 4T, 90-cm magnet(model 4T-900, Oxford Magnet Technology, Oxfordshire,UK), a clinical gradient system (model Sonata, Siemens,Erlangen, Germany), and an imaging spectrometer console(model Unity Inova, Varian, Palo Alto, CA).

To measure the PSF in the time-encoded dimension, aphantom study was performed with single-shot RASERusing a quadrature transmit/receive surface coil (twoloops, diameters � 12 cm). The phantom consisted of a600 mL beaker containing water and a 5 cm Teflon cubewith flat surfaces. The top of the cube was oriented parallelto the desired slice plane, with the cube rotated so thein-slice square cross-section had edges at �15° angles tothe axes. The phantom produces in-plane slanted edgeinterfaces suitable for PSF analysis using a modified ISO12233 measurement (9).

Human brain images were acquired using a TEM headcoil (10). Human studies were performed according to

procedures approved by the Institutional Review Board ofthe University of Minnesota Medical School after obtain-ing informed written consent.

To reduce the potential for flip angle errors, spatiallyselective refocusing was accomplished with a pair of hy-perbolic secant (HS) pulses (11) in all RASER acquisitions,which was possible because the sequence requires two180° pulses (12). Multislice RASER images were acquiredusing a 90° chirp pulse (Tp � 24 ms, and R � 225), with ablipped Gte during Tacq. Other parameters were: thk �0.5 cm, �x � 8 cm, number of slices � 10, matrix size �128 � 128, FOV � 23 � 23 cm, sw � 100 kHz, TE � 74 ms,and TR � 2.5 sec (total time � 5.5 min). Single-shotRASER images were acquired using 32 contiguous pixelsin the time-encoding dimension, with a continuous Gte

during Tacq. For comparison, single-shot spin-echo EPIimages (13,14) were also acquired using 4-ms sinc pulses.Both sequences used matrix size � 64 � 32, FOV � 22 �11 cm, TE � 65 ms, sw � 150 kHz, and thk � 5 mm. Imageswith different R values were acquired in the phantomstudy using FOV � 12 � 4 cm. The optimum RF powersetting of each type of pulse (chirp, HS, and sinc) wasdetermined empirically. This tune-up procedure involvedstepping the transmitter power until the echo amplitudereached a maximum, using a 10-sec interval between ex-citations. Scout images were acquired using a magnetiza-tion-prepared inversion-recovery sequence (turboFLASH(15)) with inversion time � 1.2 sec, TE � 3 ms, TR � 8 ms,matrix size � 128 � 128, and thk � 0.5 cm.

RESULTS

The phantom experiment was carried out to verify thenumerically determined expression for the PSF (Eq. [4]) inthe time-encoded dimension of a single-shot RASER im-age. A Gaussian-shaped profile of the PSF was obtainedusing simulations and experiments (Fig. 3b). However, thewidth of the PSF obtained by fitting experimental data wasslightly broader than that obtained by simulation (Fig. 3c),which did not include susceptibility and alignment errors.

Although RASER can be used as a single-shot singleslice imaging method, multislice RASER imaging is alsoexpected to offer advantages for certain types of experi-ments. The ability to reduce the SAR in multislice acqui-sitions will be beneficial at high fields, since SAR oftenlimits the acquisition of a large number of slices. To dem-onstrate the multislice capability of RASER, Fig. 4 showshuman brain images acquired at 4T. The images displaypredominately T2-weighted contrast, with no signs ofghosts or susceptibility artifacts. In addition, flow artifacts,which are often present in standard multislice images ac-quired without presaturation or flow compensation, werenot observed in the RASER images, as expected.

In a fast acquisition mode, RASER is expected to offersome advantages over spin-echo EPI, particularly whenfield (B0) inhomogeneity is high. Figure 5a,b shows turbo-FLASH images used to place a slice in the orbital-frontalregion of the human brain, which is commonly a difficultregion to image with EPI due to large susceptibility gradi-ents. Images were acquired to evaluate the relative qualityof “zoomed” single-shot RASER (Fig. 5d,f) versus“zoomed” spin echo EPI (Fig. 5c,e). Sets of images ac-quired with and without shimming were subtracted and

RASER: New Ultrafast MRI Method 797

the percent signal change in each pixel was calculated(Fig. 5g,h). These results demonstrate how single-shotRASER can offer improved image quality over spin echoEPI. Similar results were also obtained in imaging compar-isons performed in other areas of brain. In general, RASERappeared to be less sensitive to B0 inhomogeneity andyielded similar or better signal-to-noise ratio (SNR) thanspin echo EPI in single-shot experiments.

DISCUSSION

The initial studies reported here give an indication ofRASER’s ability to produce single-shot or multislice im-

ages with T2 contrast. Although it is anticipated thatRASER will find its greatest role as an ultrafast MRI tech-nique, multislice RASER might offer the only way to ac-complish spin-echo acquisition of large numbers of slicesat very high magnetic field (4, 7, and 9.4T), where the useof standard multislice spin-echo sequences is sometimesexcluded due to excessive SAR. Finally, RASER is a dou-ble spin-echo sequence and, therefore, refocusing can beperformed with two HS pulses to avoid flip angle errors,and under appropriate conditions, to minimize peak RFpower (12,16). RASER is relatively simple to implementon existing equipment, and it may provide advantagesover existing single-shot imaging methods for functionalimaging, especially at very high magnetic field.

Potential factors explaining the slightly increased PSFwidth in the phantom images include: limited SNR, suscep-tibility mismatch between the water and the Teflon block,and a slight tilt of the Teflon block in the through-planedirection. Despite these limitations, the agreement betweenthe experimental and theoretical PSF was reasonable. Theexperimental Gaussian fit PSF width of 1.33 pixels obtainedwith R � 550 is similar to the PSF width of 1.21 pixelsobtained with conventional phase encoding (17).

Single-shot RASER may not be suitable for human stud-ies requiring high duty-cycle (i.e., short TR) due to SARlimitations. Of the three RF pulses in RASER, the chirppulse, with its broad bandwidth, contributes the greatestSAR. As with any RF pulse, the SAR of the chirp pulseincreases in proportion with B1 squared and, therefore,SAR decreases rapidly as the flip angle is reduced. In somefuture applications it might be possible to operate RASERusing a low excitation flip angle.

The chirp pulse sequentially excites isochromats basedon a frequency sweep; consequently, a chemical shift willresult in a spatial shift in the time-encoded direction, butthis shift will be small due to the large bandwidth of thechirp pulse typically used in RASER. For example, thechemical shift between the water and methylene protons is3.4 ppm, which at 4T corresponds to a 0.8 pixel shift in thetime-encoded direction when using a chirp pulse with Tp �25 ms and R � 550. Such pixel shifts will slightly blur imagesof objects containing a wide range of chemical shifts.

Another limitation of RASER is the long TE required toobtain only low-resolution images. Future work will ex-

FIG. 4. Multislice images from a human brain acquired with multislice RASER (a–j). The images demonstrate that the quadratic phase hasa localizing effect that can be used for multislice imaging. According to the SAR monitor at the console, 2.28 W of RF power was deliveredto the coil.

FIG. 5. Axial (a) and sagittal (b) turboFLASH images of a humanbrain used to position the zoomed single-shot spin-echo EPI andRASER images. Zoomed single-shot spin-echo EPI images ac-quired without (c) and with (e) shims. Zoomed single-shot RASERimages acquired without (d) and with (f) shims. The percent changebetween the images with and without shims for spin-echo EPI (g)and RASER (h).

798 Chamberlain et al.

plore the possibility of decreasing TE by exploiting amulti-banded chirp pulse and using parallel imaging re-constructions in the time-encoded direction. Sampling onthe ramps of the readout gradient, which is currently notimplemented, is another approach that will lead to reduc-tion in TE and/or increase resolution in the time-encodeddimension in a given TE.

The RASER technique is a close relative of other time-encoding methods, but for T2-weighted imaging it has no-table advantages over its predecessor MRI sequences (2–4). The original time-encoding spin-echo sequence byMeyerand and Wong (2) suffers from contrast and signalvariation across the time-encoded dimension because asingle 180° pulse is used to generate spin-echoes, follow-ing multislice excitation by a train of 90° pulses. Moreover,in that sequence, fast gradient switching is needed be-tween 90° pulses. RASER uses only three RF pulses for anynumber of echoes, while the selection gradient (Gte) re-mains constant during the entire excitation process. Inaddition, in RASER each locally refocused echo has effec-tively the same TE, because the echoes are formed by a pairof 180° pulses (i.e., a double spin-echo). The fast imagingsequences by Frydman and colleagues (3,4) have the samedrawbacks as other imaging sequences that encode spatialinformation in a train of gradient-echoes. In these prede-cessor sequences, frequency encoding is replaced by timeencoding (also called spatial encoding), while phase en-coding is still done to encode a second spatial dimension.Like standard EPI, all echoes (except for one in the spin-echo EPI version) are gradient echoes and, thus, the apo-dizing and artifactual effects of T2*-decay are unavoidable.In single-shot RASER, on the other hand, frequency en-coding is retained, while phase encoding is replaced bytime encoding. The echoes in RASER are uniformly T2-weighted because the local refocusing plane and the vertexplane coincide and move in concert during Tacq.

RASER could also be compared with the family ofDANTE fast imaging sequences such as BURST, DUFIS,and URGE (18–20). BURST and its variants are ultrafastimaging sequences that use the sidebands of the DANTEtrain to delineate pixels in the frequency-encoded direc-tion, whereas in RASER there are no measurable side-bands. In BURST sequences an echo is formed and imagereconstruction requires FT in each dimension, whereas inRASER no FT is performed in the time-encoded dimen-sion. For a given pixel in BURST, the DANTE train excitesa strip that is narrower than the pixel width. Several ap-proaches have been proposed to improve the SNR ofDANTE-based imaging methods (21–23). One method inparticular uses a frequency-modulated (FM) DANTE (21).Although FM-DANTE offers an improvement over theoriginal BURST sequence in terms of SNR, intrapixel(voxel) phase cancellation occurs due to the quadraticphase of the excitation pulse. In RASER the magnetizationexperiences a uniform 90°, or any other desired flip angle,across the object, and in the time-encoded direction Mxy isuniform (i.e., no gaps between pixels).

In summary, RASER offers an alternative new approachto achieve T2-weighted contrast in ultrafast and multisliceMRI. Advantages of RASER over spin echo EPI includetrue T2 weighting for all echoes, reduced susceptibilityartifacts, no Nyquist ghosts, and no blurring from T2*decay between the echoes.

ACKNOWLEDGMENTSThe authors thank Drs. Djaudat Idiyatullin, Steen Moeller,and Ute Goerke for helpful discussions over the course ofthis work.

REFERENCES1. Constable RT, Gore JC. The loss of small objects in variable TE imaging:

implications for FSE, RARE, and EPI. Magn Reson Med 1992;26:9–24.2. Meyerand ME, Wong EC. A time encoding method for single-shot

imaging. Magn Reson Med 1995;34:618–622.3. Tal A, Frydman L. Spatial encoding and the single-scan acquisition of

high definition MR images in inhomogeneous fields. J Magn Reson2006;182:179–194.

4. Shrot Y, Frydman L. Spatially encoded NMR and the acquisition of 2Dmagnetic resonance images with a single scan. J Magn Reson 2005;172:179–190.

5. Dunand J-J, Delayre J. Impulse response magnetic resonance spectrom-eter. U.S. patent 3,975,675, 1976.

6. Pipe JG. Spatial encoding and reconstruction in MRI with quadraticphase profiles. Magn Reson Med 1995;33:24–33.

7. Park JY, DeleBarre L, Garwood M. Improved gradient-echo 3D magneticresonance imaging using pseudo-echoes created by frequency-sweptpulses. Magn Reson Med 2006;55:848–857.

8. Urkowitz H. Signal theory and random processes. Dedham, MA: ArtechHouse; 1983.

9. ISO 12233:2000 Photography. Electronic still picture cameras. Resolu-tion measurements.

10. Vaughan JT, Hetherington HP, Otu JO, Pan JW, Pohost GM. Highfrequency volume coils for clinical NMR imaging and spectroscopy.Magn Reson Med 1994;32:206–218.

11. Silver MS, Joseph RI, Hoult DI. Highly selective �/2 and � pulsegeneration. J Magn Reson 1984;59:347–351.

12. Conolly S, Glover G, Nishimura D, Macovski A. A reduced powerselective adiabatic spin-echo pulse sequence. Magn Reson Med 1991;18:28–38.

13. Yacoub E, Duong TQ, Van De Moortele PF, Lindquist M, Adriany G, KimSG, Ugurbil K, Hu X. Spin-echo fMRI in humans using high spatialresolutions and high magnetic fields. Magn Reson Med 2003;49:655–664.

14. Duong TQ, Yacoub E, Adriany G, Hu X, Ugurbil K, Vaughan JT, MerkleH, Kim SG. High-resolution, spin-echo BOLD, and CBF fMRI at 4 and 7T. Magn Reson Med 2002;48:589–593.

15. Haase A. Snapshot FLASH MRI: application to T1, T2, and chemicalshift imaging. Magn Reson Med 1990;13:77–89.

16. Garwood M, DelaBarre L. The return of the frequency sweep: designingadiabatic pulses for contemporary NMR. J Magn Reson 2001;153:155–177.

17. Webb AG. Optimizing the point spread function in phase-encodedmagnetic resonance microscopy. Concepts Magn Reson, Part A 2004;22A:25–36.

18. Hennig J, Hodapp M. Burst imaging. MAGMA 1993;1:39–48.19. Lowe IJ, Wysong RE. DANTE ultrafast imaging sequence (DUFIS). J

Magn Reson B 1993;101:106–109.20. Heid O, Deimling M, Huk WJ. Ultra-rapid gradient echo imaging. Magn

Reson Med 1995;33:143–149.21. Cho ZH, Ro YM, Hong IK. FM DANTE fast imaging and variations:

Emerging rf-based ultrafast imaging techniques. Concepts Magn Reson1998;10:33–54.

22. Duyn JH, van Gelderen P, Liu G, Moonen CTW. Fast volume scanningwith frequency-shifted BURST MRI. Magn Reson Med 1994;32:429–432.

23. van Gelderen P, Duyn JH, Moonen CTW. Analytical solution for phasemodulation in BURST imaging with optimum sensitivity. J Magn Reson1995;107:78–82.

RASER: New Ultrafast MRI Method 799