Synopsis
A time shifted Fast Spin
Echo (FSE) thermometry sequence which is insensitive to B0 inhomogeneity is
used in conjunction with restricted FOV module and parallel imaging to reduce
scan time and increase the number of slices. Up to 5 slices are acquired in 3
sec. This sequence works well on a clinical Focused Ultrasound brain system. The
temperature signal to noise ratio (TSNR) of this FSE sequence is approximately
2 times higher than the conventional Gradient Echo (GRE) sequence used today. 3
to 5 slices are acquired in 3 sec, rather than 1 slice with GRE.Introduction
MR‑guided Focused Ultrasound (MRgFUS) is used to
treat non-invasively many brain disorders. Currently gradient echo (GRE) is used
for thermometry imaging using the PRF shift. Recently we presented a Fast Spin
Echo (FSE) thermometry method (1) which is insensitive to B
0 inhomogeneity. In
this abstract we present improvements to this method such that it
fulfils/exceeds imaging requirements and demonstrate an advantage over GRE.
Background
The MRgFUS system consists of a large semi-spherical
helmet filled with circulating cooling water. The receive coil is a two
water-proof loops immersed in the water bath. The GRE sequence is constrained
to TE of 12.4 msec with minimum scan time and spatial resolution of 3.3 sec and
2 mm respectively with a single slice. The FSE sequence (Fig. 1) is a FSE with
the 90° pulse shifted by T sec. Two shots are acquired with
phase cycling (1). A PRF frequency shift Δf translates to a phase shift ΔΦ according
to Eq. [1]:
$$\Delta\Phi (FSE) = 4\pi \cdot \Delta f \cdot T; \space\space\space\space \Delta\Phi(GRE)=2\pi\cdot \Delta f\cdot TE \space\space\space\space\space\space\space\space [1]$$
Temperature SNR (TSNR) of FSE is higher because
1) T can be increased without affecting scan time and 2) FSE is twice more
sensitive than GRE for a given Δf, TE and T.
Method
To
enable a long echo train we modulate the RF flip angles along the echo train
(2). ky lines are acquired in a centric
order, i.e. for echo 1, 2, 3, 4, 5 … etc.
ky values are 0, 1, -1, 2, -2, 3, -3 … etc. To avoid image blurring the
echo train length (ETL) is limited to 50 echoes. During reconstruction the even
and odd echoes are separated (1) and the phase ΔΦ between them
(Eq. [1]) is determined. The simulated even/odd echo amplitude vs. echo number
is shown in Fig. 2a. The signal oscillations generate artifacts over many
echoes. We use an algorithm (beyond abstract scope) to remove the oscillation, leaving
a smooth decay (Fig. 2b). Data acquisition starts at echo 2 with minimal decay
during the echo train. Optimal TSNR is obtained at T ≈ T2*. We use T = 25 msec
which is much longer than the 12.4 msec TE in GRE.
Reduced
Field of View: To reduce scan time and
shorten echo train length (reduce blurring), we use a small FOV along the phase
direction because the treated area requires an FOV of about 10 cm. To prevent
aliasing the signal beyond the reduced FOV is saturated. The signal saturation
module is insensitive to B1 and B0 inhomogeneity. Details are described in
another abstract. By reducing the number of ky lines a full image is acquired
with a single echo train of 25 to 50 echoes and scan time of 2*TR. 3 to 5
slices are acquired in 3 sec with a TR of 1200 to 1500 msec.
Parallel Imaging (PI): Further speedup is possible with PI, but because we
have only 2 coils acceleration R is limited to 1 < R < 2. For parallel
imaging we use GRAPPA recon (TGRAPPA (3)). Fig. 3 demonstrates k-space ordering
for R = 2, where the ky lines are divided into blocks of 2 lines arranged in a
centric order. In time frame 1 line 1 in each block is acquired, in time-frame
2, line 2. Similarly, any acceleration R = p/q can be obtained by using blocks
of p lines and acquiring q lines in a block. Best results were obtained for p =
3 and q = 2, i.e. R = 3/2.
Results
We
have acquired successfully many thermal images on a brain MRgFUS system at
1.5T. Fig. 4 shows temperature data before, during and after heating from 45
time-frames. Parameters: 128x192 matrix, FOV = 24 cm, reduced FOV factor = 0.38,
ETL = 50 echoes, with 3.0 sec/frame and 3 slices. T = 25 msec.
To
compare TSNR between GRE and FSE accurately we acquired GRE and FSE images on a
conventional 1.5T scanner with a 4 channel cardiac coil. A phase shift ΔΦ was created by shifting the center frequency by Δf
= 15 Hz. TSNR is the ratio between the mean and standard deviation of the phase
images. The results are listed in Table 1.
Conclusion
The results in Table 1 indicate a 1.9: 1 TSNR
advantage of FSE over GRE with the same voxel size. FSE scans are faster with 3
to 5 slices in each time-frame. Reduced FOV provides any desired spatial
resolution with significant speedup. FSE temperature imaging works well on a
real MRgFUS system.
Acknowledgements
Goes to Alex Volovick for help with the focused ultrasound experiments.References
(1) Y. Zur, Proceedings, ISMRM 2015, p. 4054.
(2) R. F. Busse et al. Effects of refocusing flip angle modulation and view ordering in 3D Fast Spin Echo, Mag. Res. Med. 2008; 60: 640 - 649.
(3) F. A. Breuer et al., Dynamic autocalibrated parallel imaging using temporal GRAPPA (TGRAPPA), Mag. Res. Med. 2005; 53: 981 - 985.