Fat-suppressed steady-state free precession imaging using phase detection



Fully refocused steady-state free precession (SSFP) is a rapid, efficient imaging sequence that can provide diagnostically useful image contrast. In SSFP, the signal is refocused midway between excitation pulses, much like in a spin-echo experiment. However, in SSFP, the phase of the refocused spins alternates for each resonant frequency interval equal to the reciprocal of the sequence repetition time (TR). Appropriate selection of the TR results in a 180° phase difference between lipid and water signals. This phase difference can be used for fat–water separation in SSFP without any increase in scan time. The technique is shown to produce excellent non-contrast-enhanced, flow-independent angiograms of the peripheral vasculature. Magn Reson Med 50:210–213, 2003. © 2003 Wiley-Liss, Inc.

Recent advances in gradient amplifier technology have enabled the use of fully refocused steady-state free precession (SSFP) imaging methods (1, 2). SSFP imaging (or true fast imaging with steady precession (TrueFISP), balanced-fast field encoding (FFE), and fast imaging employing steady-state excitation (FIESTA)) is a rapid, efficient method that can provide high signal-to-noise ratio (SNR), good tissue contrast, and high resolution.

Recently, several methods have been proposed for SSFP imaging with fat suppression, which provides the necessary contrast between water and lipid. Fluctuating-equilibrium MR (3), suppression of lipids by RF-modulated FIESTA (4), and linear-combination SSFP (5) generate steady-state spectral profiles that suppress the frequency band containing lipid tissue. Multipoint Dixon techniques have also been combined with SSFP for fat–water separation (6). However, all of the above techniques require at least twice the acquisition time of standard SSFP. Alternatively, magnetization-prepared SSFP methods (7, 8) manipulate magnetization into the steady state after a fat-presaturation pulse. These methods are only slightly slower than standard SSFP, but they can result in transient image artifacts, particularly from transient lipid magnetization (9, 10).

This work presents a phase-sensitive SSFP technique that provides fat suppression without additional complexity or scan time in a standard SSFP sequence. The SSFP signal is refocused halfway between radiofrequency (RF) pulses, with the signal phase alternating as a function of resonant frequency. Appropriate selection of the repetition time (TR) and the center frequency results in water and lipid signals having opposite signs. By simply choosing the positive or negative signal, water-only or lipid-only images can be generated. The robust fat–water separation provided by this method is demonstrated for flow-independent peripheral angiography.


Standard SSFP imaging (1, 2) consists of RF excitation pulses spaced apart by the sequence TR, as shown in Fig. 1. All imaging gradients are rewound, and the low spatial frequency-information is acquired at an echo time (TE) halfway between RF excitation pulses. The sign of the RF pulses alternates so that high signal is obtained for on-resonance spins (11), assuming that a reasonable flip angle (α) is used. The sign of the acquisition usually alternates with the RF so that the alternating sign of the signal is removed before reconstruction.

Figure 1.

A standard SSFP pulse sequence consists of excitation pulses spaced TR apart, which alternate in sign (+α, −α, +α…). A symmetric readout is used with TE = TR/2, and all imaging gradients are rewound over a repetition. The different line segments and circles in the RF pulse correspond to different intervals, for which the transverse magnetization is shown in Fig. 2c.

Figure 2.

a: SSFP signal magnitude for TE = TR/2 = 2.3 ms and a 60° flip angle for three different T1/T2 combinations: T1/T2 = 1000/200 ms (solid line), T1/T2 = 1000/100 ms (dotted line), and T1/T2 = 200/60 ms (dashed line) correspond approximately to arterial blood, venous blood, and lipid. b: The signal phase is flat and relatively independent of T1 and T2, but alternates by π radians at intervals of 1/TR. c: The path of the magnetization over an interval of 2TR is shown for four different resonant frequencies with line styles corresponding to the RF waveform in FIG. 1. The amount of clockwise precession per TR (Δϕ) decreases (right to left) to the point (below –220 Hz) where the magnetization precesses more than a full rotation between excitations. Solid black and gray lines show flips of +αy and −αy, respectively, while precession following each tip is shown by black and gray dashed lines, respectively. Black and gray circles represent magnetization at each echo, which is always along the mx-axis. The phase that alternates with odd and even cycles is removed. However, the relative phase between lipid and water spins is still π radians.

The steady-state signal that arises after many repetitions is shown in Fig. 2a (1, 12). The signal magnitude is a strong function of the resonant frequency, with critical points spaced apart by 1/TR. At very low flip angles, these critical points are signal maxima. At typical imaging flip angles, the critical points represent signal nulls. The magnitude also varies for different relaxation times (T1 and T2), as is typical for MRI sequences.

The signal phase halfway between RF pulses in SSFP (i.e., when TE = TR/2—not when TE = 0) is, to a good approximation, a square-wave function of resonant frequency, as shown in Fig. 2b. At each critical point, the phase changes rapidly by π radians, but between critical points the phase is very flat. This means that over a frequency band of width 1/TR, the magnetization is refocused in a manner similar to a spin-echo sequence (13, 14), except that the direction of the “echo” alternates in adjacent bands. This phase characteristic of SSFP is completely independent of flip angle.

Figure 2c shows the time-course of the transverse magnetization in the mx-my plane. First the +α pulse (solid black line) tips magnetization into the half-plane with positive mx. After precession over TR (dashed black line), the −α pulse (solid gray line) tips magnetization back to the –mx half-plane. Over the next TR, the magnetization precesses back to the starting point (dashed gray line), and the cycle repeats. The precession is symmetric about the mx-axis, so that at TE = TR/2, the my component is zero. However, note that depending on the precession frequency, the mx component at the odd echo is either positive or negative. The alternating phase between even and odd echoes is removed by modulating the sign of the acquisition.

Since water and lipid have different resonant frequencies, their signals will have opposite phase with the appropriate choice of center frequency and TR, and can thus be separated. Although this separation is theoretically possible for many choices of TR, the selection of TR as the reciprocal of the resonant frequency difference between lipid and water (TR = 4.6 ms at 1.5 T) centers the respective peaks in successive signal pass-bands (Fig. 2).


This technique was validated using a 1.5 T, GE Signa LX scanner with CV/i gradients capable of 40 mT/m amplitude and 150 T/m/s slew rate, and a GE linear extremity transmit-receive coil (GE Medical Systems, Waukesha, WI).

Using a standard spin-warp readout with TR = 4.6 ms, TE = 2.3 ms, and a flip angle of 60°, 3D SSFP images were acquired of the lower leg of several normal volunteers. Imaging parameters included a resolution of 1 × 1 × 1 mm3 over a 38 × 12 × 12 cm3 field of view (FOV), resulting in a scan time of just 75 s. The readout direction of images was placed along the superior–inferior direction, as first moments are nulled in this direction so that the steady state is maintained for constant flow.

The images were reconstructed by first using a standard Fourier reconstruction, with care taken to ensure that the k-space locations had no shift (from readout timing delays or location of phase-encode lines). This is to ensure that the reconstruction itself does not introduce any linear phase, which can obscure the phase detection for fat–water separation.

After the standard reconstruction, a simple phase correction was applied. For each 2D slice, a “best-fit phase” for the complex pixel values was determined. The best-fit phase is the angle of the line passing through the origin for which the sum of squared perpendicular distances from each complex pixel value to the line is minimized. The data for the slice were rotated in the complex plane so that the best-fit phase was removed. This technique inherently weights high-signal points more heavily, and also uses both fat and water pixels to determine system phase. In practice, phases removed by this technique (for all slices) were within a range of about 50°.

The image pixels were then separated into water and lipid images based on the sign of the real part of the signal. The water images were displayed using a maximum-intensity projection (MIP) to highlight the vasculature.


The complex image data from a representative axial slice are scatter-plotted in Fig. 3. The points are well distributed along the real axis (mx-axis), corresponding to signal phases of 0 and π radians, showing that the echo phases are very stable.

Figure 3.

Scatter plot of complex image pixels. The phase of the image data does not deviate much from 0 or π radians.

Figure 4 shows the image magnitude and the separation of pixels into lipid and water images based on the sign of the real part of the image. The lipid image shows subcutaneous fat and bone marrow. The water image shows blood and muscle, with complete suppression of the lipid signal.

Figure 4.

The standard SSFP image (a) can easily be separated into water-only (b) and lipid-only (c) images by selecting only the pixels with positive and negative real parts, respectively. Lipid areas such as subcutaneous fat (dashed arrow) or bone marrow (dotted arrow) are retained in the fat-only image, while arterial blood (solid arrow), venous blood, and muscle are retained in the water-only image.

With the use of an MIP, the vessels in the lower leg can be very clearly depicted. Figure 5 shows the MIPs along two different angles. Images are zoomed to show a 24 × 12 cm3 FOV. The fat suppression provides very good contrast between arterial blood and both bone marrow and subcutaneous lipid. Image contrast is obtained without the use of a contrast agent. The T2-like SSFP contrast separates the arteries (bright signal) from veins (intermediate signal) and muscle (low signal).

Figure 5.

MIPs at two angles of the water-only 3D image, showing excellent depiction of the arterial vasculature of the lower leg of a normal volunteer. This 1 × 1 × 1 mm3-resolution image over a 24 × 12 cm2 FOV was obtained from a data set with the same resolution over a 38 × 12 × 12 cm3 FOV, acquired in just 75 s.


In refocused SSFP, the steady state is a combination of spin echoes and stimulated echoes (15). Until recently, the resonant-frequency dependence of the steady-state signal was the primary factor limiting the use of refocused SSFP. However, since the development of very rapid gradient systems has permitted the use of sufficiently short TRs, the banding artifact resulting from resonant precession is not a significant problem.

In this work we showed that the resonant frequency dependence of the signal phase can be exploited to provide spectral selectivity in images. The resulting technique uses a standard refocused SSFP imaging sequence in which the TE is halfway between RF excitation pulses. Although the most common trajectory for this method is simple Cartesian spin-warp imaging, it should also work for radial SSFP imaging (16, 17), or echo-planar SSFP imaging with an odd echo-train length (18–20).

The choice of TR as the reciprocal of the difference frequency between fat and water has the result that the lipid and water peaks are centered in adjacent signal bands. At 1.5 T, this means that the sequence is robust to resonant frequency shifts of ±110 Hz. The sequence TR can be reduced without loss of this robustness. However, increasing TR brings the frequency bands closer together, and reduces the range of frequency shifts that can be tolerated. At higher fields, such as 3 T, it is difficult to achieve a TR as short as 2.3 ms. If the TR is set to 3.0 ms instead, the sequence can still tolerate a ±110 Hz frequency variation.

Compared with other fat-suppressed SSFP imaging methods, this method provides fat suppression with no additional scan time. Magnetization-prepared SSFP techniques (7, 8) can result in significant transient artifacts, and increase the scan time compared to standard SSFP. Fluctuating-equilibrium MR or linear-combination SSFP (3, 5) techniques are more sensitive to off-resonance for a given TR than standard SSFP, as the spectral passbands are narrower. In addition, both of these sequences put restrictions on the choice of TR so that the spectral response suppresses fat. In contrast, our phase-sensitive SSFP technique should work effectively at any TR ≤ 4.6 ms at 1.5T, and does not increase scan time.

The primary limitation of fat–water separation based on image phase is a partial-volume effect. If fat and water spins occupy a single voxel, their signals interfere destructively. A small amount of water signal may be eliminated by a larger lipid signal from the same voxel, resulting in a voxel that appears to contain only lipid. This suggests that this method is appropriate for applications in which resolution is sufficiently high (so that fat and water are unlikely to occupy the same voxel), or in which fat–water interference does not affect the medical diagnosis.

Furthermore, if fat suppression is desired for the purpose of improving dynamic range, this method may not be appropriate. Since the DC components of fat and water add destructively, this method reduces the DC signal, which can improve dynamic range performance. However, when lipid signal dominates the imaging volume, other standard fat-suppression techniques may show better dynamic-range performance. With the development of improved receiver systems, this is not likely to be a serious limitation.

An important consideration for phase-sensitive SSFP imaging is phase induced by a spatially-varying coil sensitivity. The linear transmit-receive coils we used in this work had only minor phase effects, but other coils (such as surface or surface-phased-array coils) could induce significant phase. Although this phase could be prohibitive, correction for coil sensitivity is increasingly common in MRI when phased-array coils or parallel-imaging techniques are used. Similar corrections could be applied to this method as necessary.

Our phase-sensitive SSFP fat–water separation method is well suited to angiographic applications because it provides the desired “binary” vessel–background contrast. Moment-nulling over a sequence repetition can be extremely important, to allow a steady state to evolve for both static and moving materials. Since phase-sensitive SSFP fat–water separation is based on steady-state signal, it would be useful to use a trajectory that is moment-nulled over a sequence repetition in all three dimensions. Examples of this type of trajectory are 3D projection-reconstruction imaging (16) and the recently-proposed “hourglass” trajectory (21). Furthermore, although we have shown angiographic images without contrast enhancement, the fat–water separation provided by this technique could be very useful for contrast-enhanced SSFP imaging.


The combination of a phase-sensitive reconstruction with standard SSFP imaging offers a new, rapid, fat-suppressed imaging technique. The technique exploits the refocusing properties as well as the phase discontinuities observed in SSFP to provide robust fat–water separation without increasing scan time or sensitivity to off-resonance. For applications in which fat and water voxels are generally separate, such as angiography, this technique is very efficient and provides excellent image contrast.