Water-fat separation in spiral magnetic resonance fingerprinting for high temporal resolution tissue relaxation time quantification in muscle

Abstract

Purpose: To minimize the known biases introduced by fat in rapid T₁ and T₂ quantification in muscle using a single-run magnetic resonance fingerprinting (MRF) water-fat separation sequence.

Methods: The single-run MRF acquisition uses an alternating in-phase/out-of-phase TE pattern to achieve water-fat separation based on a 2-point DIXON method. Conjugate phase reconstruction and fat deblurring were applied to correct for B₀ inhomogeneities and chemical shift blurring. Water and fat signals were matched to the on-resonance MRF dictionary. The method was first tested in a multicompartment phantom. To test whether the approach is capable of measuring small in vivo dynamic changes in relaxation times, experiments were run in 9 healthy volunteers; parameter values were compared with and without water-fat separation during muscle recovery after plantar flexion exercise.

Results: Phantom results show the robustness of the water-fat resolving MRF approach to undersampling. Parameter maps in volunteers show a significant (P < .01) increase in T₁ (105 ± 94 ms) and decrease in T₂ (14 ± 6 ms) when using water-fat-separated MRF, suggesting improved parameter quantification by reducing the well-known biases introduced by fat. Exercise results showed smooth T₁ and T₂ recovery curves.

Conclusion: Water-fat separation using conjugate phase reconstruction is possible within a single-run MRF scan. This technique can be used to rapidly map relaxation times in studies requiring dynamic scanning, in which the presence of fat is problematic.

Keywords: conjugate phase reconstruction; exercise; fat; magnetic resonance fingerprinting; muscle; spiral.

Figure 1

Single‐run water‐fat resolving magnetic resonance fingerprinting (MRF) sequence and image processing pipeline. A, The MRF train, similar to the one used in Sommer et al,54 consists of 1000 flip angles, but each 2 consecutive ones having the same value being followed by a different TE. This sequence was constructed by interleaving 2 identical flip angle trains of length 500, each having its own constant TE. The entire train is preceded by an inversion pulse seen at shot number 0. B, The 1000 MRF frames are corrected for B ₀ inhomogeneities by applying the conjugate phase reconstruction (CPR) using the measured B ₀ map and the simulated spiral k‐space trajectory as input. Subsequently, the water signal is separated from the fat signal with a 2‐point DIXON, using a 7‐peak fat model. The resulting 500 fat MRF frames are deblurred by applying CPR at a stationary frequency of 440 Hz (chemical shift offset). Water and fat (WF) MRF frames are matched individually to the same dictionary using the measured $B_1^{+}$ map as input, resulting in a T₁, T₂, and M₀ map for water and for fat separately, which are combined into water and fat fraction (FF) maps

Figure 2

Water–fat‐resolved MRF parameter maps in a phantom. A, The T₁, T₂, and M₀ maps in an interleaved fully sampled experiment are shown for the water and fat part separately. Low‐signal regions after separation were masked out in the maps. Vial numbering is shown in the M₀ map. B, The T₁, T₂, and M₀ maps from an interleaved undersampled (R = 20) experiment are of very similar quality compared with those resulting from an interleaved fully sampled experiment. C, The parameter maps in (A) and (B) are both very similar to those obtained from a noninterleaved undersampled (R = 20) experiment, in which 2 separate scans were performed, each with a constant but different TE (2.3 and 3.45 ms). The noninterleaved approach shows smaller inhomogeneities compared with the interleaved approach, which may in part be explained by the longer temporal dimension of the time‐domain signals in the first case

Figure 3

Magnetic resonance fingerprinting and reference measurements in a phantom. A, T₁, T₂, and M₀ maps obtained from an interleaved, undersampled (R = 20), spiral MRF acquisition. B, Water T₁ and M₀ maps obtained from a fat‐suppressed inversion recovery (IR), water T₂ maps obtained from a multiple spin‐echo (MSE) sequence with a tri‐exponential fit, and a water and fat fraction (F) map obtained from DIXON (all Cartesian). The water T₁ maps and the water and fat fraction maps obtained with MRF are close to that obtained with fat‐suppressed IR and DIXON. The water T₂ maps obtained with the MSE sequence show shorter values compared with those obtained from the MRF measurements. The fat T₁, T₂, and M₀ maps are not shown for the standard methods, as fat suppression was used in the acquisition (IR) or during data processing (MSE)

Figure 4

Water–fat‐resolved parameter maps in a volunteer at rest. A, The T₁, T₂, and M₀ maps in a fully sampled experiment are shown for the water and the fat part separately. Low‐signal regions after separation were masked out in the maps. B, The T₁, T₂, and M₀ maps from an undersampled (R = 20) experiment are of very similar quality compared with those resulting from a fully sampled experiment. C, The parameter maps in (A) and (B) are both very similar to those obtained from an undersampled (R = 20) experiment, in which 2 separate scans were performed, each with a constant TE (2.3 and 3.45 ms). Note that results in (A)‐(C) are all obtained from separate scans, in which any type of motion may have had different effects

Figure 5

Comparison with reference measurements in a volunteer at rest. A, T₁, T₂, and M₀ maps obtained from an interleaved undersampled (R = 20) spiral MRF acquisition. B, Water T₁ and M₀ maps obtained from a fat‐suppressed IR, water T₂ maps obtained from an MSE sequence with a tri‐exponential fit, and water and fat fraction (F) maps obtained from DIXON (all Cartesian). The water T₁ map obtained with MRF is close to that obtained with fat‐suppressed IR. The T₁ map obtained with IR shows a bright region, for which the fat‐suppression pulse was probably not fully effective. The water T₂ maps obtained with the MSE sequence show shorter values compared with those obtained from the MRF measurements. Note that the fat‐suppressed reference measurements do not deliver information about the fat

Figure 6

Water‐fat resolved MRF parameter maps in a volunteer's calf. The T₁, T₂, and M₀ maps in an undersampled (R = 20) experiment are shown with separation (water, fat) and without separation for the out‐of‐phase TE (water + fat) in a volunteer at rest. The percentage difference between the water maps and the water + fat maps for all volunteers indicate that by separating the fat signal from the water signal, the mean estimated T₁/T₂ values in a region of interest in the gastrocnemius medialis muscle are significantly increased/reduced by 105 ± 94/14 ± 6 ms, underlining that fat is a confounding factor in the quantification. Low‐signal regions were masked out in the T₁ and T₂ maps

Figure 7

Water T₁ and T₂ MRF maps in a volunteer's calf before and after exercise. The water T₁ and the T₂ maps show an increase of approximately 65 ms and 9 ms, respectively, directly after exercise. The percentage difference between the parameter maps before and after exercise (with respect to at rest) shows that this increase is most pronounced in the gastrocnemius medialis (GM), the gastrocnemius lateralis (GL), and the peroneus longus (PL), whereas water T₁ and T₂ values in other muscles are mostly unchanged. Circular flow artifacts are visible around the larger vessels

Figure 8

The MRF T₁ and T₂ measurements with and without separation and reference T₂ measurements before and after exercise. A, The recovery curve of water T₂ (in milliseconds) from MRF (blue) and MSE measurements (red) averaged over a region of interest in the GM in 1 volunteer. There is a systematic difference between the water T₂ values measured with the 2 techniques, but the offset is more or less constant over time. Hence, the recovery curve measured with MRF follows the same trend as the curve measured with the reference protocol. The dashed line indicates the period during which exercise was performed. B,C, The MRF measurements with and without water–fat separation before and after exercise. The recovery curves of T₁ (B) and T₂ (C) (in milliseconds) averaged over a region of interest in the GM in another volunteer. With water–fat separation (blue), T₁ increases by approximately 35 ms and T₂ decreases by approximately 6 ms compared to without water–fat separation (red). These changes are observed along the entire curves, showing the systematic error in the presence of fat. The recovery curves show smooth behavior. The dashed line indicates the period during which exercise was performed