Variable Density Compressed Sensing Single Shot Fast Spin Echo
Valentina Taviani1, Daniel V. Litwiller2, Jonathan I. Tamir3, Andreas M. Loening1, Brian A. Hargreaves1, and Shreyas S. Vasanawala1

1Stanford University, Stanford, CA, United States, 2Global MR Applications and Workflow, GE Healthcare, New York, NY, United States, 3University of California Berkeley, Berkeley, CA, United States


Variable density (VD) sampling was implemented into an extended echo train single shot fast spin echo (SSFSE) pulse sequence. Compressed sensing (CS) reconstuction was used. With respect to regular undersampling and ARC (Autocalibrated Reconstruction for Cartesian imaging), VD CS SSFSE allows higher acceleration factors, which translates in increased flexibility in the choice of echo times for full-Fourier imaging (shorter minimum TEs) and faster acquisitions (shorter breath-holds).


Single-shot fast spin echo (SSFSE) is being increasingly used for pelvic and abdominal imaging due to its robustness to motion and excellent T2 contrast. Recent work has shown that variable refocusing flip (VRF) angles along the echo train can reduce SAR (Specific Absorption Rate) and therefore scan time, as well as blurring due to T2 decay1. In particular, the T2 prolongation effect due to VRF allows full k-space coverage, which results in higher SNR and reduced blurring when compared to conventional half-Fourier SSFSE. While clinically relevant echo times have been demonstrated with full-Fourier coverage for specific applications1, the use of higher resolution and other constraints on the flip angle schedule (e.g. to limit signal loss due to cardiac pulsation in abdominal imaging) often lead to suboptimal echo times. Parallel imaging is routinely used to overcome this limitation; however, noise amplification and residual aliasing limit practical acceleration factors. Here we show that variable density (VD) sampling and compressed sensing (CS) can be used to extend the range of clinically relevant echo times achievable with full-Fourier k-space coverage, while maintaining image quality (IQ) and further increasing scanning efficiency.


A VD sampling scheme was implemented into an extended echo train SSFSE pulse sequence (Fig.1). Ns views (out of a fully-sampled set) were mapped to a nonlinear function to produce the VD pattern shown in Fig.1, with an effective acceleration factor R=Ns/N, where Ns is the number of acquired phase-encoding lines and N represents the number of lines for Nyquist-rate sampling. This resulted in low frequencies being sampled more densely than higher frequencies, which is a more efficient sampling strategy given the signal evolution produced by the VRF modulation, especially at lower flip angles (cfr. Fig. 1c). A slightly different sampling pattern was obtained by imposing higher undersampling in the first half of k-space, prior to reaching the center of k-space, to further reduce the minimum achievable TE (Fig.1a). L1-ESPIRiT2 calibration and image reconstruction with L1-wavelet regularization were performed using the BART package3. For comparison, regularly undersampled datasets with the same net acceleration and number of autocalibrating lines were acquired and reconstructed using ARC (Autocalibrated Reconstruction for Cartesian imaging4). In all cases, half-sinc excitation and full-Fourier k-space coverage were used. The VRF schedule was controlled by prescribing first, minimum and last flip angles as well as the flip angle corresponding to the center of k-space. A 90° minimum flip angle was used to minimize signal loss due to cardiac pulsation in the left lobe of the liver. Imaging was performed at 3T (GE MR750, Waukesha, WI) using the 20 upper elements of a 32-channel receive-only torso coil (NeoCoil, Pewaukee, WI) with R/L phase-encoding (4 coil elements across phase FOV).

Results and Discussion

For moderate acceleration factors (~3.5 in a phantom, ~2.5 in vivo), VD and CS gave similar IQ as regular undersampling with ARC reconstruction when the same net acceleration and number of autocalibrating lines were used (Fig. 2a and 3). Note that in both Fig. 2a and 3, VD and CS gave less residual aliasing and noise amplification than regular sampling with ARC. In a phantom, VD enabled undersampling factors up to 4.5, which resulted in a 25% minimum TE reduction (Fig.2). In vivo, regular sampling resulted in residual aliasing and noise amplification when the acceleration factor exceeded 2.5. Conversely, VD allowed net accelerations up to 3.3, which reduced minimum TE up to 24% with respect to regular undersampling (Fig.4). Minimum TR reductions between 15 and 20% were observed, with corresponding scan time reductions and reduced breath-hold durations, with no significant impact on IQ. VD sampling did not result in significantly sharper images, probably due to the relatively high minimum flip angle used (cfr. Fig.1c). The asymmetric sampling pattern gave a slightly lower $$$\sqrt{R}$$$ penalty, although IQ was similar to that obtained with symmetric sampling at the highest undersampling level. For heavily T2-weighted imaging, asymmetric VD and CS can allow higher resolution than regular undersampling, provided high enough SNR can be obtained (Fig.5). In its current implementation, CS reconstruction takes less than a minute per slice. The addition of coil compression would reduce reconstruction times by approximately a factor of 4.


We have introduced VD sampling and CS image reconstruction to SSFSE and achieved higher acceleration factors than with regular undersampling and ARC. This can be used to realize shorter echo times while avoiding half-Fourier sampling. Optimization of the VD sampling pattern in relation to the signal evolution produced by the VRF schedule could further reduce T2-induced blurring.


GE Healthcare, NIH P41-EB015891-18 and R01-EB009690-1.


[1] Loening AM, Saranathan M, Ruangwattanapaisarn N, Litwiller DV, Shimakawa A, Vasanawala SS. Increased speed and image quality in single-shot fast spin echo imaging via variable refocusing flip angles. J Magn Reson Imaging 2015; Epub ahead of print.

[2] Uecker M, Lai P, Murphy MJ, Virtue P, Elad M, Pauly J, Vasanawala SS, Lustig M. ESPIRiT - An Eigenvalue Approach to Autocalibrating Parallel MRI: Where SENSE meets GRAPPA. Magn Reson Med 2014; 71:990-1001.

[3] Uecker M, Ong F, Tamir JI, Bahri D, Virtue P, Cheng JY, Zhang T, Lustig M. Berkeley Advanced Reconstruction Toolbox. Proc. Intl. Soc. Mag. Reson. Med. 2015; 23:2486.

[4] Brau AC, Beatty PJ, Skare S, Bammer R. Comparison of reconstruction accuracy and efficiency among autocalibrating data-driven parallel imaging methods. Magn Reson Med 2008; 59:382-95.


Regular, VD and asymmetric VD sampling (A). Note incoherent aliasing introduced by the VD pseudo-random sampling pattern (B). In principle, VD can sample the VRF-generated signal more efficiently, especially when low minimum refocusing flip angles are used, resulting in almost constant signal levels towards the center of k-space (C).

VD CS SSFSE gives less residual aliasing and noise amplification than regular undersampling, effectively allowing higher acceleration factors and up to 25% shorter echo times.

Regular undersampling with ARC (A) and VD CS SSFSE (B) give similar image quality and nearly identical TE (143ms) and TR (592/597ms) for the same number of phase-encoding (R=2.5) and autocalibrating (N=20) lines. Other imaging parameters common to both acquisitions were: 40cm FOV, 320×224 matrix, 5mm slice thickness, 83.3kHz bandwidth.

2.5x regular undersampling (A) vs. 3.3x VD CS SSFSE (B). Shorter TEs and TRs were obtained with VD CS SSFSE. Note improved visualization of liver parenchyma at shorter echo times (same window level between A and B). In all cases, a 320×224 matrix, 5mm slice thickness and 83.3kHz bandwidth were used. FOV ranged between 30 and 48cm.

(A) 320×224 matrix with 2.5x regular sampling vs. (B) 320×320 matrix with 2.5x asymmetric VD sampling (40cm FOV, 5mm slice thickness, 83.3kHz bandwidth). Asymmetric VD allowed higher resolution (30% smaller voxels) while preserving image quality. Note increased sharpness of liver capsule (yellow arrows) and pancreatic duct (white arrows).

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)