Dual-Echo Interleaved Echo-Planar Imaging of the Brain
`
`Glenn S. Slavin, Kim Butts, John N. Rydberg, Clifford R. Jack, Stephen J. Riederer
`
`An interleaved echo-planar imaging (EPI) technique is de-
`scribed that provides images from 20 sections of the brain at
`two echo times (27 and 84 ms) in 1:M. Six echoes per image
`per repetition are collected in 24 repetitions of the pulse
`sequence. MR images of the brain obtained from five volun-
`teers using the dual-echo EPI sequence, fast spin-echo (FSE),
`and conventional dual-echo spin-echo were evaluated quali-
`tatively for diagnostic use and quantitatively for relative sig-
`nal-to-noise ratio (SNR), contrast, and contrast-to-noise
`ratios (CNR).
`Key words: echo-planar imaging; brain; MR; fast MR imaging.
`
`INTRODUCTION
`As introduced over 10 years ago, the two-dimensional
`Fourier-transform multi-slice spin-echo
`technique
`continues to be the standard procedure for clinical
`magnetic resonance imaging (MRI) of the brain (1).
`This is due to the high quality of the images in terms of
`signal-to-noise ratio and tissue contrast. Furthermore,
`conventional spin-echo (SE) is capable of providing
`both early and late echo images from one acquisition.
`The importance of acquiring long TR short TE images
`lies in the ability to provide pathology/cerebrospinal
`fluid (CSF) contrast as a complement to the pathology/
`gray matterlwhite matter contrast inherent in long TR
`long TE (T,-weighted) images. Nevertheless, it is well
`recognized that these high quality images are obtained
`at the cost of a relatively long scan time, typically on
`the order of 8 min for 256 X 192 resolution images
`using a 2500-ms TR. The purpose of this work was to
`develop a fast scan technique for providing long TR
`multiple echo images from a large volume such as the
`brain within an acquisition time arbitrarily selected to
`be 1 min long. The expectation is that such a sequence
`could be useful in providing MR images of patients for
`whom long acquisition times are problematic.
`A number of fast imaging techniques have been devel-
`oped in an attempt to reduce acquisition time while
`maintaining the overall quality of conventional T,-
`weighted spin-echo (T,SE) imaging. These include vari-
`ants of rapid acquisition with relaxation enhancement
`(RARE) imaging (2,3), combined gradient- and spin-echo
`(GRASE) imaging (4), and echo-planar imaging (EPI) (5).
`Hybrid RARE, or “fast” or “turbo” spin-echo (FSE),
`uses multiple repetitions of trains of individually phase-
`
`MRM 33264-270 (1995)
`From the Magnetic Resonance Laboratory, Department of Diagnostic Ra-
`diology, Mayo Clinic, Rochester, Minnesota
`Address correspondence to: S. J. Riederer, Ph.D., MR Laboratory, Mayo
`Clinic, 200 1st Street SW, Rochester, MN 55905.
`Received August 5, 1994; revised October 26. 1994; accepted October 26,
`1994.
`This work was supported in part by NIH Grant R01 CA37993 and by General
`Electric Medical Systems.
`0740-3194/95 $3.00
`Copyright 0 1995 by Williams & Wilkins
`All rights of reproduction in any form reserved.
`
`encoded RF-refocused spin-echoes. Although it is theo-
`retically possible to reduce the T,SE acquisition time by
`a factor equal to the number of echoes acquired per
`repetition, increasing the length of the echo trains can
`decrease the number of sections imaged per TR. Dual-
`echo contrast can be achieved in several ways. First, with
`appropriate phase-encoding assignments, the early and
`late echoes can be acquired from two separate scans.
`This, however, defeats some of the advantage of time
`reduction provided by the RARE technique. Alterna-
`tively, a single scan can be devised in which the first half
`of the collected data is incorporated into the early-echo
`image and the second half of the data is incorporated into
`the late-echo image (6). However, this method also limits
`the number of sections available for imaging relative to
`conventional SE, thereby compromising reduced scan
`time to achieve coverage of the targeted volume. A mod-
`ified version of FSE has been developed to generate dual-
`echo images by view sharing (7). In clinical studies, this
`method has been used to produce dual-echo images at 40
`and 80 ms of 18 sections in 2 min. Because some of the
`measurements are used in more than one image, the
`shared-echo approach in general has better efficiency for
`image formation at multiple echo times than a standard
`FSE approach.
`Although FSE imaging of the brain has in general dem-
`onstrated contrast and overall image quality similar to
`conventional SE imaging (8, 9), it has been shown to be
`subject to several complications. These include blurring
`of early echo or spin density-weighted images (lo), edge
`enhancement and ringing in T,-weighted images, and
`perhaps most significantly, decreased sensitivity to mag-
`netic susceptibility effects (9). Furthermore, FSE appears
`more prone to magnetization transfer effects than does
`conventional SE (11).
`GRASE imaging, which utilizes short gradient-recalled
`echo trains centered about each of a number of RF-refo-
`cused spin-echoes, has also been proposed as a means to
`reduce imaging time while maintaining SE contrast.
`While GRASE appears to produce contrast comparable
`with conventional SE, because it is closely related to
`FSE, it too can be penalized in terms of fewer imaged
`sections compared with conventional SE. Recently, a
`variant of GRASE has been presented (12) in which im-
`ages corresponding to two echo times are generated for
`each section. With this technique as described, only the
`late echo is generated in the true GRASE fashion; with
`only one refocusing pulse, the early echo is acquired in a
`manner identical to that of interleaved EPI and similar to
`that presented in this work.
`As originally described, EPI collects the entire raw
`image data set by sampling the free induction decay after
`a single excitation, although variants can be used (13)
`with a refocusing pulse to provide later echo times as
`desired for T,-weighting. Recently, dual-echo single-shot
`EPI, using three 90° pulses to generate a spin-echo
`
`264
`
`General Electric Co. 1035 - Page 1
`
`

`
`Dual-Echo Interleaved EPI of the Brain
`
`265
`
`(TI31 = 50-80 ms) and a stimulated echo (TE2 = 200-
`500 ms) has been demonstrated (14). Such single-shot
`methods, which can generate images in as little as tens of
`milliseconds, necessitate sampling bandwidths of sev-
`eral hundred kHz. In addition to the reduced signal-to-
`noise ratio (SNR) resulting from the high bandwidth,
`single-shot EPI is also susceptible to resolution limita-
`tions and off-resonance artifacts as a consequence of the
`extended readout duration (15). These resolution and
`off-resonance limitations can potentially be addressed
`with specialized gradient systems.
`In an attempt to exploit the EPI methodology while
`simultaneously addressing some of its intrinsic limita-
`tions, several methods of “interleaved” or “multi-shot”
`EPI acquisition have been explored (16-21). These meth-
`ods acquire multiple gradient-refocused echoes in each
`of several repetitions. In fact, one such variant (19, 21)
`has recently been shown to provide diagnostic quality
`images of the liver with acquisition times of less than 20
`s. Given these results and the considerations described
`above, the specific purpose of this work was to design a
`multi-slice dual-echo interleaved EPI technique for rapid
`MR imaging of the brain.
`
`METHODS
`Motivation
`To facilitate a comparison between the fast-scan se-
`quences discussed in the Introduction and the dual-echo
`EPI technique presented here, a summary of pulse se-
`quence characteristics is presented in Table 1. It should
`be emphasized that the values presented in this table are
`simply representative of those used in research or clini-
`
`Table 1
`Summary of Fast-Scan Pulse Sequence Characteristics
`
`cal settings. A certain degree of freedom is available to
`the user in choosing values for most of the listed param-
`eters. These choices will inevitably affect the number of
`imaged sections as well as the total acquisition time.
`The first two columns in Table 1 list the pulse se-
`quences and their corresponding references. The first
`entry, conventional dual-echo SE, is considered the stan-
`dard technique against which the others are ultimately
`compared. The third column lists the duration of data
`acquisition within one repetition. For techniques that
`involve repeated 180’ pulses (e.g., FSE and GRASE), this
`duration is considered to extend from the start of the first
`echo to the end of the last echo. For techniques that
`utilize gradient-recalled echo trains with only one 180’
`pulse per TE (e.g., EPI, SE), the duration of data acquisi-
`tion is the duration of the echo train. Column 5 shows a
`second basis for comparison. The inter-echo time is ef-
`fectively the duration of data acquisition (column 3) di-
`vided by the number of echoes collected (column 4)
`during that time. It can be interpreted as one measure of
`a technique’s efficiency of data collection. The values
`listed for number of echoes per repetition per image
`(column 4) sometimes called the echo train length, or
`ETL, as well as the TR (column 7) are taken either from
`the cited reference or from clinical protocols. The acqui-
`sition times in column 8 are those resulting from the
`aforementioned parameters. Two values are shown for
`each technique: those not in parentheses for equal fields-
`of-view (FOV) in the readout and phase-encoding direc-
`tions and those in parentheses if a three-fourths FOV is
`used in the phase-encoding direction (usually the pa-
`tient’s right/left).
`
`1
`
`2
`
`Technique
`
`Reference
`
`Standard
`Dual-echo SE
`
`One image per section
`Single-shot EPI
`Multi-shot EPI
`Single-echo GFIASE
`Single-echo FSE
`
`(1 1
`
`(5)
`(19, 21)
`(4)
`(2)
`
`3
`
`4
`
`z2::: Echoes per
`
`repetition
`acquisition per image
`(msec)
`8 + 16b
`
`1
`
`5
`
`6
`
`7
`
`Inter-echo
`time
`(msec)
`
`Bandwidth Sections imaged
`per scan
`(256 X 192)
`(kH.4
`
`8
`
`~
`
`9
`
`Acquisition ST lr
`
`time per
`scana
`
`dual-echo
`sectionsa
`
`516, ?ab
`
`21 (TR = 2300)
`
`7:49 (552)
`
`7:49 (5x52)
`
`100
`80
`188
`127
`
`128
`16
`24
`8
`
`0.25-1 .o
`5
`7.8
`17
`
`250-500
`232
`232
`216
`
`arbitraw
`12 (TR = 2000)
`13 (TR = 2500)
`15 (m = 2600)
`
`100mdimage
`(0:18)‘
`0:23 (0:18)
`1 :08 (052)
`
`(1 :03)‘
`1:12 (056)”
`3:21 (2:34)‘
`
`(1 4)
`
`73
`
`4
`4
`7,9b
`6
`
`1.5
`
`17
`20
`2.6, 9.ab
`5
`
`Two images per section
`110 + 110
`Dual-echo single-shot
`EPI
`59 + 59
`(6)
`Dual-echo FSE
`128
`(7)
`Shared-echo FSE
`26 + 88
`(1 2)
`Dual-echo GFIASE
`30 + 30
`Dual-echo interleaved This work
`EPI
`See text for details. The technique proposed here, dual-echo interleaved EPI, is the last entry in the table.
`a Acquisition times for the 3/4 FOV case are shown in parentheses.
`Values for early and late echoes, respectively.
`‘ One acquisition for the early echo plus one acquisition for the late echo (7R = 3400 ms). Also, employs 3/4 k-space sampling.
`128 x 128 resolution.
`‘ One acquisition ( W T R = 4/2200) for the early echo plus one acquisition (ETUTR = 8/3500) for the late echo.
`One acquisition for the early echo plus one acquisition for the late echo (7R = 4000 ms).
`g Employs 1/2 k-space sampling.
`rR = 3000 ms.
`
`255
`
`arbitravg
`
`300md2 img
`
`216
`216
`50, 216.7b
`? 32
`
`20 (TR = 3500)
`18 (TR = 2600)
`20 (TR = 3500)
`20 (TR = 2600)
`
`2 5 5 (2:13)
`2:55 (2:13)
`2:30 (1 :54)h
`2:lO (1 :38)
`1:41 (1:17)
`1:41 (1:17)
`1 :26 (1 :05) 1 :26 (1 :05)
`
`General Electric Co. 1035 - Page 2
`
`

`
`266
`
`As stated previously, the purpose of this work was to
`develop an RF-refocused interleaved EPI pulse sequence
`that could provide both early and late echo images, e.g.,
`at 30 and 80 ms echo times, for multiple sections from a
`single 1-min imaging acquisition. Specific design consid-
`erations included the ability to provide high resolution,
`heavily T,-weighted, multiple-section images of the
`brain. It was felt that these requirements could be met by
`using 256 X 192 resolution and a repetition time greater
`than 2000 ms. The first echo time (TE1) was targeted to
`be short enough that CSF would be no brighter than the
`surrounding brain parenchyma. The second echo time
`(TE2) was targeted to be sufficiently long that both CSF
`and typical pathology would be bright compared with
`brain parenchyma but not so long that SNR would be
`severely compromised. To provide adequate image qual-
`ity, additional RF and gradient waveforms were incorpo-
`rated for both fat suppression as well as spatial presatu-
`ration of flowing spins. The ultimate goal was to use the
`above specifications to image the entire brain in a single
`acquisition. From our experience, adequate coverage can
`be achieved by imaging 20 sections with a 5-mm thick-
`ness and 2.5-mm gap. The sequence was implemented on
`a standard 1.5 T GE Signa system with 1 G/cm maximum
`gradient amplitudes and 600 ps rise time from zero to full
`scale.
`The last column in Table 1 gives approximations for
`the total scan time necessary to generate dual-echo im-
`ages from at least 20 sections. These scan times are a
`consequence of selecting parameters such as ETL and TR
`which allow appropriate weighting, adequate coverage,
`and acceptable quality of the early and late echo images
`while attempting to minimize the acquisition period.
`These times are illustrative and may be slightly longer or
`shorter depending on the actual choice of parameters.
`Given the specific design considerations, it was con-
`cluded that the dual-echo interleaved EPI sequence, the
`last entry in Table 1, was a plausible candidate.
`
`Pulse Sequence
`The dual-echo EPI pulse sequence is shown in Fig. 1.
`Data acquisition periods in the form of gradient-recalled
`echo trains are placed around each of the two RF-refo-
`
`-
`
`G,
`
`RF
`
`I
`80
`
`I
`100
`
`I
`0
`
`I
`2 0
`
`I
`40
`
`I
`6 0
`Time (rnsec)
`FIG. 1. Dual-echo interleaved EPI pulse sequence diagram. Posi-
`tions of the sliding echo trains for the 1st and 12th of the 24
`waveform are depicted with solid and
`repetitions of the G,,
`dotted lines, respectively. Although not shown, the phase-encod-
`ing blips of the Gphase waveform slide in synchrony with their
`respective echo trains.
`
`Slavin et al.
`
`cused echo times. By targeting echo times near 30 and 80
`ms, the maximum duration available for each gradient
`echo train is approximately 40 ms. This estimate is used
`in determining the maximum possible number of echoes
`per echo train, i.e., the echo train length (ETL). With a
`receiver bandwidth of -+32 kHz, the read window dura-
`tion for an individual 256-point echo is about 4 ms.
`Including ramp times, the inter-echo time is 5 ms. Addi-
`tionally, the technique of “sliding echo trains” (17, 21-
`24) is incorporated into both acquisition periods to
`smooth out magnitude and phase transitions in the raw
`data. Because the total sliding period is approximately
`equal to the duration of a single read window, each of the
`two data acquisition periods must allow room for ETL+ 1
`echoes. As a consequence of these constraints, an ETL of
`six, which provides two 30-ms data acquisition periods,
`was selected. Therefore, to achieve 256 X 192 resolution,
`32 repetitions of the sequence are required, with six
`echoes acquired per image per repetition. For this work,
`a three-quarters FOV in the phase-encoding direction
`was used, permitting the number of repetitions to be
`reduced to 24.
`The phase-encoding procedure is illustrated in Fig. 2,
`again using the following parameters: y-resolution of 192,
`ETL of 6, and 24 repetitions (144 total views). After the
`90’ excitation and first 180’ refocusing pulses, six lines
`of data (corresponding to views 1,25,49,73,97, and 121)
`are acquired as in conventional interleaved EPI for the
`early echo image (Fig. 2a). The second 180’ pulse inverts
`the position along both k, and k,, and six lines of data
`(corresponding to views 24, 48, 72, 96,120, and 144) are
`acquired for the late echo image (Fig. 2b). Thus, k, is
`traversed in the same direction, top to bottom, for both
`echo trains. During subsequent repetitions, however, the
`echo train for the first data set must step in the -k,
`direction (downward in Fig. 2a), while the echo train for
`the second data set must step in the +k, direction (up-
`ward in Fig. 2b). For example, on the second repetition,
`views 2,26,50, 74,98, and 122 are collected for the early
`echo image while views 23,47, 71, 95, 119, and 143 are
`collected for the late echo image.
`The echo time of the first image (TE1) is defined as the
`time after the 90’ excitation at which the k, = 0 line is
`measured. For RF-refocusing, this typically occurs at ex-
`actly twice the delay time between the goo and 180’
`pulses. In Fig. 2, the six groups of k, lines (24 lines per
`group) sampled by the six echoes of each echo train are
`shown to exactly straddle ky = 0. A problem with this for
`the bandwidth and resolution used in this work is that
`measuring three groups of ky lines prior to refocusing
`forced TE1 to be no smaller than 37 ms. This echo time
`resulted in images with unacceptably high CSF signal
`relative to surrounding brain parenchyma. Accordingly,
`to reduce TE1, fewer ky lines were measured prior to the
`refocusing time, resulting in asymmetric filling of
`k-space along the k, direction. For this work, k, = 0 was
`straddled by the second group of k, lines. Specifically,
`phase-encodings +AkJ2 and -AkJ2
`are acquired on
`views 36 and 37 for the early echo and views 108 and 109
`for the late echo, where Ak,, is the sampling interval used
`along k,.
`
`General Electric Co. 1035 - Page 3
`
`

`
`Dual-Echo Interleaved EPI of the Brain
`
`267
`
`semndmu
`
`Rep 1
`Rep 2
`Rep 3
`
`I
`4 ......................................................................
`
`----------__ - - - - - - _ _ _ _ _ _ _ kx 73+72 4 i
`- - - _ _ _ _ _ _ _ _ _ _
`----_-__---_
`I
`,
`
`- ....................................................................
`I
`b k x
`
`t
`
`a.
`
`t
`
`b.
`
`FIG. 2. Schematic of k-space tra-
`jectories for dual-echo interleaved
`EPI. Three of the 24 repetitions are
`shown. Each repetition after data
`collection ends for the first echo (0
`in part (a)), the readout position is
`inverted along both k, and k,, by the
`second 180” pulse. This new posi-
`tion determines where data collec-
`tion begins for the second echo (0
`in part (b)).
`
`To implement the sliding echo trains with this phase-
`encoding scheme, on successive repetitions the first echo
`train is preceded by a small cumulative time delay while
`the second echo train is provided an equally cumulative
`time advance. In Fig. 1, this corresponds to the simulta-
`neous sliding of the first and second echo trains to the
`right and left, respectively. As mentioned previously, the
`purpose of the sliding echo trains is to allow continuous
`sampling of the T,* decay curves, thereby providing
`smooth magnitude and phase modulations over k-space.
`The difference in the sliding directions compensates for
`filling k-space “backwards” in the second image.
`
`In Vivo Imaging
`Multi-section MR images of the brain of five healthy
`volunteers were obtained using the dual-echo EPI se-
`quence and compared with FSE and SE images of the
`same sections. The dual-echo EPI sequence was per-
`formed with a TElITE2ITR of 27/84/2500 and 256 X 192
`resolution. Axial sections, 5 mm thick with 2.5 mm sep-
`aration, were acquired using a 22-cm FOV and a 314 FOV
`in the phase-encoding direction. Two dummy repeti-
`tions, played out prior to the 24 repetitions of data col-
`lection, were included to set the spins close to a steady
`state. Fat saturation was used to reduce the intensity of
`chemical shift misregistration (15), while inferior spatial
`presaturation (25) and first-order gradient moment null-
`
`ing (26) on the late echo in the slice-select direction were
`used to minimize artifacts due to flowing blood and CSF.
`For comparison, FSE and conventional dual-echo SE
`were performed with parameters substantially matching
`those of the EPI (Table 2). The FSE used two acquisitions
`with different echo train lengths at a 516 lcHz band-
`width: TE/TRIETL of 17/2500/4 for the early echo and
`85/2500/8 for the late echo. The SE sequence used a
`TElITE2ITR of 27/84/2500 with variable bandwidth:
`216 lcHz for the first echo and +8 lcHz for the second.
`Both FSE and SE utilized inferior spatial presaturation,
`5-mm sections with 2.5-mm separation, and 22-cm FOV
`with 314 FOV in the phase-encoding direction.
`Dual-echo images from 20 sections were obtained us-
`ing EPI, FSE, and SE in 1:05,3:15, and 6:30, respectively.
`Although the FSE technique with a 17-ms TE is typi-
`cally used at our institution for the early echo, the FSE
`was also performed using a 34-ms TE, as this echo time
`more closely matched the TE1 used for SE and EPI.
`
`Evaluation
`Images were evaluated based on signal-to-noise (SNR],
`contrast and contrast-to-noise ratio (CNR) calculations
`and overall image quality as determined using radiolog-
`ical criteria. Mean signal measurements, from three sec-
`tions per technique, were made in uniform regions of
`both gray and white matter. The regions of interest were
`
`Table 2
`Scan Parameters
`
`Sequence
`
`SE
`
`FSE
`
`TEITR
`2712500
`8412500
`
`1712500
`8512500
`3412500
`
`ETL
`
`1
`1
`
`4
`8
`4
`
`BW
`(W
`216
`-C8
`
`216
`216
`216
`
`Sections
`
`Imaging options
`
`Sat-I
`Sat-I
`
`Sat-I
`Sat-I
`Sat-I
`
`20
`2oa
`20
`
`EPI
`
`232
`6
`2712500
`Sat-I, Fat Sat
`232
`8412500
`6
`Sat-I, Fat Sat, GMN
`NOTE: Three-fourths FOV in the phase-encoding direction and 5 mm section thickness are used for all techniques. Sat-I = inferior spatial presaturation; Fat
`Sat = spectral saturation of fat; GMN = first order gradient moment nulling in slice-select direction.
`a Two acquisitions were necessary to achieve 20 sections at this TR.
`This Scan was performed to facilitate comparison with the first echo of the SE and EPI scans.
`
`Acquisition time
`
`} 6:30
`] 3:15
`1 :35
`1 :40b
`} 1:05
`
`General Electric Co. 1035 - Page 4
`
`

`
`268
`
`chosen to be in identical locations for both early and late
`echo images for each of the three techniques. Standard
`deviations of the noise were measured in regions of the
`image outside the head along the phase-encoding direc-
`tion, excluding areas corrupted by pulsatile flow artifact.
`Gray matter SNR was defined as gray matter signal di-
`vided by the standard deviation of the noise, and white
`matter SNR was defined similarly using the white matter
`signal. Gray mattedwhite matter contrast was defined as
`the gray matter signal minus the white matter signal
`divided by the white matter signal. Contrast-to-noise ra-
`tios were calculated as the signal difference between gray
`and white matter divided by the standard deviation of
`the noise. For each technique, mean contrast and CNR
`with associated standard deviations were determined by
`averaging the contrast and CNR values from the five
`volunteers. The results were then normalized to those
`from the SE images. Image quality was assessed radiolog-
`ically on the basis of anatomic resolution and contrast as
`well as prominence of any artifact.
`
`RESULTS
`Early and late echo images of the SE (a and b), FSE (c and
`d), and EPI (e and fl sequences hom one section of one
`volunteer are shown in Fig. 3.
`Table 3 shows the results of SNR, contrast, and CNR
`measurements for the three techniques. The EPI and
`FSE SNR measurements for gray and white matter ap-
`proximately scale with bandwidth relative to SE. Both
`EPI and FSE images exhibit significantly higher aver-
`age gray matterlwhite matter (GM/WM) contrast on the
`early echo, by a factor exceeding 1.50, relative to con-
`ventional SE. EPI, FSE, and SE all show similar average
`
`Slavin et al.
`
`contrast on the second echo. Average CNR from the
`early EPI echo is similar to the SE values due primarily
`to the increased contrast. On the other hand, average
`CNR from the late EPI echo is lower, by a factor of
`0.52 2 0.10, relative to SE.
`Radiologically, the images in all studies were felt to
`have adequate contrast characteristics in the early and
`late echoes, as targeted. Several EPI images showed arti-
`fact due to magnetic susceptibility effects near regions of
`air-tissue interfaces. These artifacts occurred predomi-
`nantly at sinuses and at the skull base but also appeared
`in isolated locations near the surface of the brain. In
`addition, ghosting artifacts due to pulsatile flow of both
`blood and CSF seemed to be more apparent with EPI than
`with FSE or SE. Once again, these artifacts were most
`prominent at the more inferior locations in the head.
`Clearly, many of these radiological criteria can only be
`better assessed in prospective patient studies.
`
`DISCUSSION
`This work was undertaken to develop an interleaved EPI
`pulse sequence aimed at providing clinically useful dual-
`echo images of the whole brain within a 1-min scan time.
`We feel that the results of this study demonstrate that
`early and late echo images obtained with the new dual-
`echo EPI sequence are of promising quality and, due to
`the short acquisition time, are especially suited for pa-
`tients uncooperative for standard imaging procedures.
`Results from a preliminary clinical study performed with
`the dual-echo EPI sequence are shown in Fig. 4.
`The EPI results generated in this work show G M I W
`contrast similar to FSE in both echoes and significantly
`higher than SE in the early echo. The cause of this in-
`
`FIG. 3. Early and late echo images
`from one section of the head of a
`healthy volunteer. (a, b) Conven-
`tional spin-echo. (c, d) Fast spin-
`echo. (e, 9 EPI.
`
`General Electric Co. 1035 - Page 5
`
`

`
`Dual-Echo Interleaved EPI of the Brain
`
`269
`
`Table 3
`Relative SNR and Gray MatterMlhite Matter (GMMIM) Contrast and CNR for EPI and FSE versus SE
`Relative
`white matter
`SNR
`1 .o
`1 .o
`
`Sequence
`
`TEITR
`
`SE
`
`27/2500
`84/2500
`
`Bandwidth
`(kH4
`
`216
`28
`
`Relative gray
`matter S N R
`1 .o
`1 .o
`
`Relative GM/
`WM contrast
`1 .o
`1 .o
`
`FSE
`
`17/2500
`34/2500
`85/2500
`
`216
`216
`216
`
`EPI
`
`232
`27/2500
`2 32
`84/2500
`Measurements are averages over five volunteers.
`
`1.02 2 0.03
`0.93 2 0.02
`0.79 2 0.04
`
`0.65 2 0.02
`0.46 2 0.04
`
`0.97 2 0.03
`0.88 2 0.02
`0.80 2 0.04
`
`0.62 2 0.02
`0.45 2 0.03
`
`1.53 2 0.16
`1.64 2 0.02
`0.96 2 0.16
`
`1.53 2 0.20
`1.15? 0.17
`
`Relative GM/
`WM CNR
`1 .o
`1 .o
`1.44 * 0.04
`
`1.48 2 0.11
`
`0.76 2 0.13
`
`0.95 2 0.12
`0.52 2 0.10
`
`severe SNR penalty is on the second echo, where EPI
`uses a ?32 kHz bandwidth compared with 2 8 kHz for
`conventional SE. This fourfold increase in bandwidth
`results in a twofold reduction in SNR. It is conceivable
`that this problem of SNR loss with the EPI sequence can
`be addressed with improved receiver coils.
`Magnetic susceptibility effects can manifest them-
`selves as several types of artifacts. Because data is col-
`lected along a train of gradient-recalled echoes, signal
`loss due to spin dephasing can be expected in regions of
`susceptibility differences. Furthermore, due to the rela-
`tively long (30 ms) data acquisition periods, large phase
`differences can accumulate between tissues with differ-
`ent magnetic susceptibilities. These phase differences
`give rise to spatial shifts in the phase-encoding direction
`which can cause overlapping of structures in the image.
`The resulting artifacts appear as streaks of very high
`signal intensity. Nevertheless, the increased sensitivity
`to susceptibility effects that cause artifacts in some EPI
`sections may, in other sections, prove advantageous with
`respect to FSE. The short inter-echo spacing in FSE has
`been implicated in the reduced sensitivity of FSE to
`susceptibility variations (3, 27). Because EPI would be
`expected to have a sensitivity to magnetic susceptibility
`somewhere between that of SE and gradient-recalled
`echo imaging, iron-containing structures and hemor-
`rhagic lesions may be better depicted with EPI than with
`FSE. However, this has yet to be shown in a patient
`series.
`As suggested by column 9 of Table 1, the interleaved
`EPI sequence can be considered one sample along a
`continuum of pulse sequences which can be ordered
`according to the acquisition time required to perform
`whole-brain T,-weighted imaging. For the shortest
`times-several hundred milliseconds per section (or
`images can be
`up to 10 s for the entire brain)-the
`prone to off-resonance effects and resolution limits as
`well as reduced SNR because of the required high
`bandwidth. These effects become less severe as times
`are increased to the 40-s to 1 min range, where either
`averaging of single-shot or implementation of multi-
`shot echo-planar methods can be employed. Incorpo-
`ration of RF-refocusing for some or all echoes reduces
`the sensitivity to off-resonance but typically causes an
`increased inter-echo time and increased scan time.
`Finally, conventional spin-echo requires the longest
`
`d
`C
`FIG. 4. Dual-echo EPI (a, b) and conventional spin-echo (c, d)
`images of the same section of the head of a 38-year-old male
`diagnosed with multiple sclerosis. The patient was scanned with
`the conventional SE sequence before the EPI. The EPI used the
`parameters listed in Table 2, while the SE had the following minor
`differences: TEl/TE2/TR = 30/80/2200, 20 x 20 cm FOV, and
`first-order gradient moment nulling in the section-select direction.
`FSE was not used. Scan times for the SE and EPI were 7:29 and
`1 :05, respectively. Qualitatively, the pathology is visible (arrows) in
`the EPI images (a, b) to a degree similar to that in the SE images
`(c, d). Slight section misalignment resulted from interscan patient
`movement.
`
`crease in contrast requires further study. Despite the
`improved contrast, the EPI images demonstrate a reduced
`CNR relative to SE and FSE. Decreased SNR, which is
`reflected in CNR, is expected with the EPI sequence due
`primarily to the higher sampling bandwidth. The most
`
`General Electric Co. 1035 - Page 6
`
`

`
`270
`
`Slavin et al.
`
`acquisition time, 5 min or more, but has few of these
`limitations. The operating point along this continuum
`will likely depend on user preference.
`In summary, we have demonstrated the ability of an
`interleaved RF-refocused echo-planar pulse sequence to
`produce multiple-section dual-echo images of the whole
`brain in a 1-min long acquisition period, approximately
`six to seven times shorter than that for conventional
`spin-echo.
`
`REFERENCES
`1. M. Brant-Zawadski, D. Norman, T.H. Newton, Magnetic
`resonance imaging of the brain: the optimal screening tech-
`nique. Radiology 152, 71-77 (1984).
`2. J. Hennig, A. Naureth, H. Friedburg, RARE imaging: a fast
`imaging method for clinical MR. Magn. Reson. Med. 3,823-
`833 (1986).
`3. P. S. Melki, R. V. Mukern, L. P. Panych, F. A. Jolesz, Com-
`paring the FAISE method with conventional dual-echo se-
`quences. J. Magn. Reson. Imaging 1, 319-326 (1991).
`4. D. A. Feinberg, K. Oshio, GRASE (Gradient- and Spin-Echo)
`M R imaging: a new fast clinical imaging technique. Radiol-
`ogy 181, 597-602 (1991).
`5. P. Mansfield, Multi-planar image formation using NMR spin
`echoes. J. Phys. C10, L55-L58 (1977).
`6. P. S. Melki, F. A. Jolesz, R. V. Mulkern, Partial RF echo-
`planar imaging with the FAISE method. II. Contrast equiv-
`alence with spin-echo sequences. Magn. Reson. Med. 26,
`342-354 (1992).
`7. B. A. Johnson, E. K. Fram, B. P. Drayer, B. L. Dean, P. J.
`Keller, R. Jacobowitz, Evaluation of shared-view acquisition
`using repeated echoes (SHARE): a dual-echo fast spin-echo
`MR technique. AJNR 15,667-673 (1994).
`8. K. M. Jones, R. V. Mulkern, M. T. Mantello, P. S. Melki, S. S.
`Ahn, P. D. Barnes, F. A. Jolesz, Brain hemorrhage: evalua-
`tion with fast spin-echo and conventional spin-echo images.
`Radiology 182(1), 53-58 (1992).
`9. A. M. Norbash, G. H. Glover, D. R. Enzmann, Intracerebral
`lesion contrast with spin-echo and fast spin-echo pulse se-
`quences. Radiology 185, 661-665 (1992).
`10. R. V. Mulkern, S . T. S. Wong, C. Winalski, F. A. Jolesz, Con-
`trast manipulation and artifact assessment of 2D and 3D
`RARE sequences. Magn. Reson. Imaging 8, 557-566 (1990).
`11. R. T. Constable, A. W. Anderson, J. Zhong, J. C. Gore, Fac-
`tors influencing contrast in. fast spin echo M R imaging.
`Magn. Reson. Imaging 10,497-511 (1992).
`12. D. A. Feinberg, B. Kiefer, A. W. Litt, Dual contrast GRASE
`
`(gradient-spin echo) imaging using mixed bandwidths.
`Magn. Reson. Med. 31(4), 461-464 (1994).
`13. M. S. Cohen, R. M. Weisskoff, Ultra-fast imaging. Magn. Re-
`son. Imaging 9, 1-37 (1991).
`14. P. Bornert, D. Jense