Downloaded from www.ajronline.org by 177.8.149.40 on 10/17/15 from IP address 177.8.149.40. Copyright ARRS. For personal use only; all rights reserved
951
Review
Fast MR Imaging: E. M. Haacke1
and
Techniques
and Clinical
J. A. Tkach2
Fast MR imaging has matured in the past few years and is now of established value for several aspects of clinical MR imaging. The initial impetus for rapid imaging was to reduce scan times. Today its usefulness includes reducing motion artifacts, improved contrast per unit time, three-dimensional (3-D) imaging, real-time imaging, cine-mode imaging, and flow imaging. The focus of this review is on short-TR steady-state gradient-echo imaging. We discuss the basic sequence design of the mainstream fast techniques. Many important applications exist, including gadopentetate dimeglumine-enhanced MR imaging of the brain and spine, subsecond imaging of real-time applications, myelographic imaging of the spine, cardiac cine-mode imaging; 3-D musculoskeletal (knee) imaging, 3-D pituitary imaging; two-dimensional and 3-D body imaging; 3-D carotid and intravascular imaging, and reformatting 3-D images into arbitrary planes.
Fast imaging has its impetus in improving throughput of patients and reducing motion artifacts. With respect to acquisition time, MR imaging can be classified into four regimens: (1) conventional scans taking minutes, as in normal SE imaging; (2) intermediate savings in time (by factors of 2 or 4) such as with hybrid imaging [1, 2], multiecho techniques [3], partial-Fourier imaging [4-8], or improved reconstruction with limited data [9-li]; (3) rapid GFE scanning with partial flip angles [12-23] requiring on the order of 1-10 sec; and (4) snapshot imaging, actually a subset of rapid GFE scanning [24, 25] (collected in less than 1 sec) and EPI [26-29] (col-
Received January 17, 1990; accepted Department of Radiology. University reprint requests 2 Department AJR 155:951-964,
Applications
after revision May 21, 1990. Hospitals of Cleveland, and the Department
to E. M. Haacke, Department of Radiology, The Cleveland
November
lecting the data in a single readout in 20-1 00 msec). The focus of this article is on rapid partial-flip-angle GFE techniques. The MR abbreviations and symbols used throughout this article are defined in the Appendix. The last few years have seen a resurgence of earlier attempts at fast, short IR techniques (known as SSFP imaging), thanks to improvements in field homogeneity, higher signal-to-noise (S/N) at higher fields, and higher gradient strengths. Much of the theoretical foundation for these techniques was laid several decades ago [1 5, 30-33]; its reincarnation in the last few years has appeared in the guise of techniques like FLASH [1 3, 1 4, 1 8, 20] and FISP [1 2, 21, 22]. Many important contributions regarding contrast-to-noise (C/N) [1 2, 34-45] and other issues such as spoiling [46, 47], slice profile [36, 44], missing pulse acquisition [48], and clinical usefulness [49-51 ] are now being carefully examined. There are also a great many ways to approach steady state and to acquire the data [1 6, 1 7, 22, 23, 48, 52-54]. In this article, the basic concept of steady state is reviewed, followed by a summary of the different Fourier-transform MR sequence types available on modem whole-body MR systems. Their physical differences are addressed, with special emphasis on the consequences for imaging parameters such as measuring time, spatial resolution, S/N, and C/N. The effects of slice profile, resonance offset variations, and local field inhomogeneities on image contrast are also mentioned. Clinical examples of 2-0 and 3-D implementations [55, 56] are then presented for CNS, musculoskeletal, cardiac, and vascular imaging.
of Physics,
of Radiology, University Hospitals of Cleveland, Clinic Foundation, Cleveland, OH 44195-5129.
1990 0361 -803X/90/1
555-0951
© American
Roentgen
Case Westem 2074 Abington
Ray Society
Reserve
University,
Rd., Cleveland,
Cleveland,
OH 44106.
OH 44106.
Address
952
HAACKE
Downloaded from www.ajronline.org by 177.8.149.40 on 10/17/15 from IP address 177.8.149.40. Copyright ARRS. For personal use only; all rights reserved
Techniques Because of recent improvements in field homogeneity, S/N, and gradient strengths, GFE imaging has become a viable means of signal acquisition. In conventional SE imaging, the minimum TA and TE are limited by the fact that two RF pulses, 90#{176} and 1 80#{176} (Fig. 1 A), are applied per cycle. Both TR and TE can be reduced when the signal is formed by gradient reversal alone. For these short-TR GFE sequences, only one RF pulse is applied per cycle (Fig. 1 B), reducing RF power deposition. With present gradient strengths of 10 mT/ m, by using a short RF pulse, small sampling intervals, and reduced resolution, it is possible to reduce TEs to 2-5 msec and TRs to 4-8 msec [24, 25, 57]. A scan with 128 phase-
AND
TKACH
AJR:155,
November
1990
encoding steps can then be acquired in roughly 0.5-i .0 sec. Reducing the number of phase-encoding steps would allow faster image acquisition at the expense of resolution. RF times can be selective at 1 -3 msec and nonselective (covering the whole volume) at 500 sec or less. Both 2-0 and 3-0 acquisitions are possible. Flip angles of less than 90#{176} are then selected to optimize contrast. This inherent contrast can be altered by the application of various preparation schemes (e.g., r saturation pulse) prior to image acquisition [24, 25]. Owing to the absence of the 180#{176} refocusing pulse in GFE sequences, only the dephasing induced by the imaging gradients (with the exception of phase-encoding) is refocused at the TE. Dephasing due to local field inhomogeneities remains, acting to accelerate the rate of decay of the transverse t.
-.411!..-
,,
c
Li
EiIiL
II
r1
rL\\
II
Li
---
II
II
:
Li
Li
Li.,,
\-
-,
B)
---= A
C
B
.L_
“
__A__
I’
,
Li11
Li
5’.
,t-
,..
-h--
-4 I
[71
U
tkt-
-\\-
I
Li
,,,,
-“
D
Li11
r1
U
-
E
Fig. 1.-A, Usual SE sequence (TI2 first RF, second RF pulse) refocuses all spins at TE, even in presence of static local field inhomogeneities. Alternatively, dephasing gradient before the pulse can be inverted and moved to after the pulse (the broken line along G,). Similarly, the polarity of the rephasing gradient pulse (as well as phase-encoding gradient) in the slice direction can be inverted and placed after the pulse. S (+) represents the signal during the pulse sequence. Phi represents the phase evolution for an isochromat in all directions. The dotted line (labeled Bz) corresponds to sliceselect gradient (G..), thin solid line (labeled By) to phase-encoding gradient (G,,), and dark solid line (labeled Bx) to read gradient (G,). Phi is shown in each sequence diagram. B, FLASH gradient-echo sequence. Unbalanced gradients and changing gradient table in slice-select direction are used as spoilers. Elongation of read gradient beyond sampling window ensures that if spoiling is incomplete, no RF echoes begin to form during data collection, as well as providing for integration over resonant offset. Both are essential to avoiding image artifacts. C, SSFP, true FISP sequence, measures total signal FID and RF echo portions superimposed. Associated sequence structure is characterized by balanced imaging gradients in all three directions (using the broken line in G,,). Rare ROAST is balanced in all but the slice-select direction (using the solid line in G.e). It could have been unbalanced in the read direction instead. D, ROAST sequence. Only gradients in phase-encoding direction are balanced. This is common version of FISP, which collects just the FID. Rare ROAST integrates the SSFP signal from a voxel across a 0-2w range of resonant offset as a consequence of unbalanced gradient structure along G,. (or G,). For ROAST, this integration occurs along both G. and G,. E, Contrast-enhanced ROAST sequence collects only RF portion of data. Two repeat intervals are shown to highlight phase behavior. As in ROAST, integration across resonant offset occurs along and G, for the sequence structure shown.
AJR:155,
November
1990
FAST
MR
Downloaded from www.ajronline.org by 177.8.149.40 on 10/17/15 from IP address 177.8.149.40. Copyright ARRS. For personal use only; all rights reserved
magnetization (referred to as T2*). This vulnerability has proved to be a limitation ofthese techniques. Alternative rapid techniques have been implemented to avoid T2*related signal and resolution loss [52, 54, 56, 57], the details of which are discussed later under imaging difficulties. SSFP SSFP implies that both longitudinal and transverse magnetization reach a dynamic equilibrium that is the same from cycle to cycle [31 ] (from one RF excitation pulse to the next). Considering only the behavior of the transverse magnetization over one cycle, the steady-state signal is characterized by an exponentially decreasing function, reaching a minimum near the center of the interval, followed by an exponentially increasing function of opposite polarity reaching its maximum just prior to the next RF pulse. The initial decaying portion of this complete SSFP signal is typically referred to as the FID and the regrowth as the RF echo. The RF echo is similar to the 90-i 80#{176} echo, in that dephasing due to local field inhomogeneities is refocused at the RF pulse. As long as the interval between RF pulses remains constant and no imaging gradients are applied (or the gradient structure is constant from cycle to cycle), all echoes will occur at a time nTR, where n is an integer. The components that refocus to form the echoes continue to undergo Ti growth and T2 decay during the intervals in which they are stored along the longitudinal axis or lie in the transverse plane, respectively. However, they are continually augmented by those echo trains initiated by subsequent RF pulses so that a steady-state signal is obtained. Echo amplitude is dependent on Ti T2, TR, flip angle (a), and precession angle or resonant offset As shown by Kaiser et al. [32], the expression for the SSFP signal can be determined by summing the appropriate refocusing echo trains, treating them as a geometric series. The result is identical to that obtained by solving the phenomenologic Bloch equations of motion for equilibrium conditions [33]. When balanced imaging gradients are applied (i.e., the gradient-induced phase accumulation before the gradientecho time, TE, of the FID is equal and opposite to that which accumulates over the TR-TE interval), all echoes occur simultaneously with the gradient echo of the FID at TE (this sequence is shown in Fig. 1 C, broken-line box in Alternatively, the gradients can be purposely unbalanced and the associated sequence designed so that only the gradient echo of the FID (Figs. 1 C and 1 D) or RF echo (Fig. 1 E) forms and is acquired at TE. Whenever a TR less than or on the order of Ti is used, it will take several RF pulses or sequence repetitions for the spin system to reach an equilibrium (where the dynamic behavior of the magnetization during any interpulse interval is identical to that during any other) to some chosen degree of accuracy. The exact number of cycles required will differ depending on Ti, T2, TR, a, /3, and the sequence structure. The approach to equilibrium can be quite long when TR or a or both are small. In the rotating frame of reference, the resonant offset rotation angle, /3, is the angle through which the spin pre,
(a).
IMAGING
953
cesses in the transverse plane from one RF pulse to the next (i.e., over one TR interval). In practice, /3 is determined by the RF phase, imaging gradients, and local field inhomogeneities [57], all of which shift the frequency off resonance. Nomenclature.-The widespread simultaneous development of many of the sequences designed to acquire the FID and RF echo portions of the SSFP signal has led to a proliferation of names. For example, the incompletely refocused (with respect to the zero moment of the imaging gradients in the slice and/or read directions) SSFP sequence designed to acquire the FID portion of the SSFP signal is called FAST, GRASS, and modified FISP (true FISP actually includes balanced gradients along all axes) [30]. Because the signal acquired with this intermediate sequence reflects an average over the resonant offset angle induced by the remnant imaging gradients and other static field inhomogeneities across a voxel, we propose to lump the three names into resonant offset averaged steady state, or ROAST, to reflect the actual spin system behavior. “Rare ROAST” is used to denote those sequences in which the gradients along only one of the imaging directions is left unbalanced. As is true for ROAST sequences (Fig. 1 0), the net effect of the imaging gradients in contrast-enhanced FAST or contrast-enhanced ROAST, also called PSIF, the sequence designed to acquire the SE portion of the SSFP signal, is nonzero in the slice or read directions or both. An SSFP sequence can also be structured to collect a gradient echo of both the FID and the RF echo simultaneously [21]. In the original article by Hahn [30], the concept of SSFP was used for conditions in which “coherence” of the transverse and longitudinal magnetization was maintained. We recommend classifying sequences such as FISP, ROAST, or CE-ROAST, in which the net effect of the imaging gradients is constant from cycle to cycle, as in SSC sequences. Sequences such as FLASH or spoiled GRASS, in which the transverse magnetization is spoiled (when TA is much less than T2 or when /3 is randomized), would then be referred to as steady-state incoherent (551) sequences. SSC behavior-When TA
951
Review
Fast MR Imaging: E. M. Haacke1
and
Techniques
and Clinical
J. A. Tkach2
Fast MR imaging has matured in the past few years and is now of established value for several aspects of clinical MR imaging. The initial impetus for rapid imaging was to reduce scan times. Today its usefulness includes reducing motion artifacts, improved contrast per unit time, three-dimensional (3-D) imaging, real-time imaging, cine-mode imaging, and flow imaging. The focus of this review is on short-TR steady-state gradient-echo imaging. We discuss the basic sequence design of the mainstream fast techniques. Many important applications exist, including gadopentetate dimeglumine-enhanced MR imaging of the brain and spine, subsecond imaging of real-time applications, myelographic imaging of the spine, cardiac cine-mode imaging; 3-D musculoskeletal (knee) imaging, 3-D pituitary imaging; two-dimensional and 3-D body imaging; 3-D carotid and intravascular imaging, and reformatting 3-D images into arbitrary planes.
Fast imaging has its impetus in improving throughput of patients and reducing motion artifacts. With respect to acquisition time, MR imaging can be classified into four regimens: (1) conventional scans taking minutes, as in normal SE imaging; (2) intermediate savings in time (by factors of 2 or 4) such as with hybrid imaging [1, 2], multiecho techniques [3], partial-Fourier imaging [4-8], or improved reconstruction with limited data [9-li]; (3) rapid GFE scanning with partial flip angles [12-23] requiring on the order of 1-10 sec; and (4) snapshot imaging, actually a subset of rapid GFE scanning [24, 25] (collected in less than 1 sec) and EPI [26-29] (col-
Received January 17, 1990; accepted Department of Radiology. University reprint requests 2 Department AJR 155:951-964,
Applications
after revision May 21, 1990. Hospitals of Cleveland, and the Department
to E. M. Haacke, Department of Radiology, The Cleveland
November
lecting the data in a single readout in 20-1 00 msec). The focus of this article is on rapid partial-flip-angle GFE techniques. The MR abbreviations and symbols used throughout this article are defined in the Appendix. The last few years have seen a resurgence of earlier attempts at fast, short IR techniques (known as SSFP imaging), thanks to improvements in field homogeneity, higher signal-to-noise (S/N) at higher fields, and higher gradient strengths. Much of the theoretical foundation for these techniques was laid several decades ago [1 5, 30-33]; its reincarnation in the last few years has appeared in the guise of techniques like FLASH [1 3, 1 4, 1 8, 20] and FISP [1 2, 21, 22]. Many important contributions regarding contrast-to-noise (C/N) [1 2, 34-45] and other issues such as spoiling [46, 47], slice profile [36, 44], missing pulse acquisition [48], and clinical usefulness [49-51 ] are now being carefully examined. There are also a great many ways to approach steady state and to acquire the data [1 6, 1 7, 22, 23, 48, 52-54]. In this article, the basic concept of steady state is reviewed, followed by a summary of the different Fourier-transform MR sequence types available on modem whole-body MR systems. Their physical differences are addressed, with special emphasis on the consequences for imaging parameters such as measuring time, spatial resolution, S/N, and C/N. The effects of slice profile, resonance offset variations, and local field inhomogeneities on image contrast are also mentioned. Clinical examples of 2-0 and 3-D implementations [55, 56] are then presented for CNS, musculoskeletal, cardiac, and vascular imaging.
of Physics,
of Radiology, University Hospitals of Cleveland, Clinic Foundation, Cleveland, OH 44195-5129.
1990 0361 -803X/90/1
555-0951
© American
Roentgen
Case Westem 2074 Abington
Ray Society
Reserve
University,
Rd., Cleveland,
Cleveland,
OH 44106.
OH 44106.
Address
952
HAACKE
Downloaded from www.ajronline.org by 177.8.149.40 on 10/17/15 from IP address 177.8.149.40. Copyright ARRS. For personal use only; all rights reserved
Techniques Because of recent improvements in field homogeneity, S/N, and gradient strengths, GFE imaging has become a viable means of signal acquisition. In conventional SE imaging, the minimum TA and TE are limited by the fact that two RF pulses, 90#{176} and 1 80#{176} (Fig. 1 A), are applied per cycle. Both TR and TE can be reduced when the signal is formed by gradient reversal alone. For these short-TR GFE sequences, only one RF pulse is applied per cycle (Fig. 1 B), reducing RF power deposition. With present gradient strengths of 10 mT/ m, by using a short RF pulse, small sampling intervals, and reduced resolution, it is possible to reduce TEs to 2-5 msec and TRs to 4-8 msec [24, 25, 57]. A scan with 128 phase-
AND
TKACH
AJR:155,
November
1990
encoding steps can then be acquired in roughly 0.5-i .0 sec. Reducing the number of phase-encoding steps would allow faster image acquisition at the expense of resolution. RF times can be selective at 1 -3 msec and nonselective (covering the whole volume) at 500 sec or less. Both 2-0 and 3-0 acquisitions are possible. Flip angles of less than 90#{176} are then selected to optimize contrast. This inherent contrast can be altered by the application of various preparation schemes (e.g., r saturation pulse) prior to image acquisition [24, 25]. Owing to the absence of the 180#{176} refocusing pulse in GFE sequences, only the dephasing induced by the imaging gradients (with the exception of phase-encoding) is refocused at the TE. Dephasing due to local field inhomogeneities remains, acting to accelerate the rate of decay of the transverse t.
-.411!..-
,,
c
Li
EiIiL
II
r1
rL\\
II
Li
---
II
II
:
Li
Li
Li.,,
\-
-,
B)
---= A
C
B
.L_
“
__A__
I’
,
Li11
Li
5’.
,t-
,..
-h--
-4 I
[71
U
tkt-
-\\-
I
Li
,,,,
-“
D
Li11
r1
U
-
E
Fig. 1.-A, Usual SE sequence (TI2 first RF, second RF pulse) refocuses all spins at TE, even in presence of static local field inhomogeneities. Alternatively, dephasing gradient before the pulse can be inverted and moved to after the pulse (the broken line along G,). Similarly, the polarity of the rephasing gradient pulse (as well as phase-encoding gradient) in the slice direction can be inverted and placed after the pulse. S (+) represents the signal during the pulse sequence. Phi represents the phase evolution for an isochromat in all directions. The dotted line (labeled Bz) corresponds to sliceselect gradient (G..), thin solid line (labeled By) to phase-encoding gradient (G,,), and dark solid line (labeled Bx) to read gradient (G,). Phi is shown in each sequence diagram. B, FLASH gradient-echo sequence. Unbalanced gradients and changing gradient table in slice-select direction are used as spoilers. Elongation of read gradient beyond sampling window ensures that if spoiling is incomplete, no RF echoes begin to form during data collection, as well as providing for integration over resonant offset. Both are essential to avoiding image artifacts. C, SSFP, true FISP sequence, measures total signal FID and RF echo portions superimposed. Associated sequence structure is characterized by balanced imaging gradients in all three directions (using the broken line in G,,). Rare ROAST is balanced in all but the slice-select direction (using the solid line in G.e). It could have been unbalanced in the read direction instead. D, ROAST sequence. Only gradients in phase-encoding direction are balanced. This is common version of FISP, which collects just the FID. Rare ROAST integrates the SSFP signal from a voxel across a 0-2w range of resonant offset as a consequence of unbalanced gradient structure along G,. (or G,). For ROAST, this integration occurs along both G. and G,. E, Contrast-enhanced ROAST sequence collects only RF portion of data. Two repeat intervals are shown to highlight phase behavior. As in ROAST, integration across resonant offset occurs along and G, for the sequence structure shown.
AJR:155,
November
1990
FAST
MR
Downloaded from www.ajronline.org by 177.8.149.40 on 10/17/15 from IP address 177.8.149.40. Copyright ARRS. For personal use only; all rights reserved
magnetization (referred to as T2*). This vulnerability has proved to be a limitation ofthese techniques. Alternative rapid techniques have been implemented to avoid T2*related signal and resolution loss [52, 54, 56, 57], the details of which are discussed later under imaging difficulties. SSFP SSFP implies that both longitudinal and transverse magnetization reach a dynamic equilibrium that is the same from cycle to cycle [31 ] (from one RF excitation pulse to the next). Considering only the behavior of the transverse magnetization over one cycle, the steady-state signal is characterized by an exponentially decreasing function, reaching a minimum near the center of the interval, followed by an exponentially increasing function of opposite polarity reaching its maximum just prior to the next RF pulse. The initial decaying portion of this complete SSFP signal is typically referred to as the FID and the regrowth as the RF echo. The RF echo is similar to the 90-i 80#{176} echo, in that dephasing due to local field inhomogeneities is refocused at the RF pulse. As long as the interval between RF pulses remains constant and no imaging gradients are applied (or the gradient structure is constant from cycle to cycle), all echoes will occur at a time nTR, where n is an integer. The components that refocus to form the echoes continue to undergo Ti growth and T2 decay during the intervals in which they are stored along the longitudinal axis or lie in the transverse plane, respectively. However, they are continually augmented by those echo trains initiated by subsequent RF pulses so that a steady-state signal is obtained. Echo amplitude is dependent on Ti T2, TR, flip angle (a), and precession angle or resonant offset As shown by Kaiser et al. [32], the expression for the SSFP signal can be determined by summing the appropriate refocusing echo trains, treating them as a geometric series. The result is identical to that obtained by solving the phenomenologic Bloch equations of motion for equilibrium conditions [33]. When balanced imaging gradients are applied (i.e., the gradient-induced phase accumulation before the gradientecho time, TE, of the FID is equal and opposite to that which accumulates over the TR-TE interval), all echoes occur simultaneously with the gradient echo of the FID at TE (this sequence is shown in Fig. 1 C, broken-line box in Alternatively, the gradients can be purposely unbalanced and the associated sequence designed so that only the gradient echo of the FID (Figs. 1 C and 1 D) or RF echo (Fig. 1 E) forms and is acquired at TE. Whenever a TR less than or on the order of Ti is used, it will take several RF pulses or sequence repetitions for the spin system to reach an equilibrium (where the dynamic behavior of the magnetization during any interpulse interval is identical to that during any other) to some chosen degree of accuracy. The exact number of cycles required will differ depending on Ti, T2, TR, a, /3, and the sequence structure. The approach to equilibrium can be quite long when TR or a or both are small. In the rotating frame of reference, the resonant offset rotation angle, /3, is the angle through which the spin pre,
(a).
IMAGING
953
cesses in the transverse plane from one RF pulse to the next (i.e., over one TR interval). In practice, /3 is determined by the RF phase, imaging gradients, and local field inhomogeneities [57], all of which shift the frequency off resonance. Nomenclature.-The widespread simultaneous development of many of the sequences designed to acquire the FID and RF echo portions of the SSFP signal has led to a proliferation of names. For example, the incompletely refocused (with respect to the zero moment of the imaging gradients in the slice and/or read directions) SSFP sequence designed to acquire the FID portion of the SSFP signal is called FAST, GRASS, and modified FISP (true FISP actually includes balanced gradients along all axes) [30]. Because the signal acquired with this intermediate sequence reflects an average over the resonant offset angle induced by the remnant imaging gradients and other static field inhomogeneities across a voxel, we propose to lump the three names into resonant offset averaged steady state, or ROAST, to reflect the actual spin system behavior. “Rare ROAST” is used to denote those sequences in which the gradients along only one of the imaging directions is left unbalanced. As is true for ROAST sequences (Fig. 1 0), the net effect of the imaging gradients in contrast-enhanced FAST or contrast-enhanced ROAST, also called PSIF, the sequence designed to acquire the SE portion of the SSFP signal, is nonzero in the slice or read directions or both. An SSFP sequence can also be structured to collect a gradient echo of both the FID and the RF echo simultaneously [21]. In the original article by Hahn [30], the concept of SSFP was used for conditions in which “coherence” of the transverse and longitudinal magnetization was maintained. We recommend classifying sequences such as FISP, ROAST, or CE-ROAST, in which the net effect of the imaging gradients is constant from cycle to cycle, as in SSC sequences. Sequences such as FLASH or spoiled GRASS, in which the transverse magnetization is spoiled (when TA is much less than T2 or when /3 is randomized), would then be referred to as steady-state incoherent (551) sequences. SSC behavior-When TA
