In this work we present a 4D radial fat-suppressed alternating-TR bSSFP sequence with T2-like contrast for abdominal imaging during free-breathing. Abdominal MR-imaging, for example in pancreatic cancer patients, is challenging as the pancreas is small, embedded in fat, and moves during breathing. Therefore, a high resolution, respiratory motion robust, and signal-to-noise ratio efficient sequence with fat saturation is highly desired. Next, good anatomical contrast by T2-weighting is a prerequisite.

General sequence design: The sequence was based on the alternating repetition time (TR) balanced steady state free precession (ATR)¹ principle, which provides T2-like contrast and fat-suppression by alternating between two TRs (TR1, TR2) and acquiring data during TR1 only. The 4D acquisition (3 spatial dimensions + time) was performed with a stack-of-stars radial readout with tiny golden angle increments². The radial acquisition facilitated a compressed sensing (CS) reconstruction with retrospective sorting of the data into images with desired spatial and temporal resolution. The sequence was self-gated owing to the continuous acquisition of radial k-lines passing through the center of k-space, which facilitated high-resolution navigator-based reconstructions of data in different breathing motion states.

Subjects: 3 healthy volunteers and 2 pancreatic cancer patients.

Sequence parameters: The sequence was implemented on a 3T Philips Ingenia scanner. Sequence parameters were: FOV=300x300x100mm³, TR1/TE=3.5/1.8ms, τ=TR1/TR2=3, and FA=17°. The acquisition was performed with 20 radial stacks with additional 1.4x slice oversampling. For each radial angle, spokes from all stacks were acquired before advancing to the next radial angle with a tiny golden angle increment of ~23.63°. The acquisition was done continuously during 3 minutes of free-breathing. For comparison, a T1-weighted turbo field echo (TFE) with TFE-factor of 28 and a SPIR fat suppression pre-pulse prior to each TFE-shot was scanned using an identical radial sequence.

Extraction of respiratory motion: Post processing and reconstruction was done in Matlab. Data were corrected for eddy currents and B0-phase delays³ . An inverse Fourier transformation was performed in the kz stack direction. These were used to calculate the Z-intensity weighted projection signal⁴, yielding the center of mass of mean slice intensity which provided the respiratory motion navigator information. A fourth-order low-pass filter was applied to remove the rotational frequency arising from the radial acquisition. The rotational frequency in k-space is n_z*TR*goldenangle/360 =0.67 Hz, where n_z is the stack size. The respiratory motion was subsequently binned in 10 equally timed motion states (figure 1).

Compressed sensing reconstruction: Sorting resulted in 10 respiratory phases containing 114 spokes for a reconstruction matrix size of 300x300x20, corresponding to a 1x1 mm² in-plane resolution. This corresponds to an undersampling factor of 4.1 times. We used the BART toolbox⁵ to perform a parallel-imaging CS reconstruction with a temporal total-variation l₁-regularization constraint r=0.01, and 150 iterations. Because of full sampling in the stack direction, the reconstruction could be performed in parallel for every slice. Reconstruction time for a full dataset was 2 hours on a server with two Intel Xeon E5‐2690 processors and 128GB RAM.

We were able to obtain 4D (10 timeframes) high resolution (1x1x5mm³) respiratory cycle images (figure 2). Adequate fat suppression, T2 contrast and details of moving structures are clearly visible.

The extent of spatial movement in the CS reconstruction is similar to that in the linear reconstruction (figure 3), suggesting that temporal smoothing was minimal.

Our method also performed well on TFE-sequences in which spins can be prepared, such as the T1-weighted TFE with a SPIR as preparation pulse (figure 4). Due to the T2-like contrast of ATR, different organs are better appreciated than in the T1-weighted TFE. Therefore we performed only the ATR sequence in patients.

We were able to perform high resolution reconstructions (1x1 mm² in-plane pixel size) for a 3 minute free-breathing acquisition in pancreatic cancer patients. The tumor and cystic/necrotic lesions are visible. This acquisition allowed for 10 respiratory phases to be reconstructed.

In this work we have introduced a 4D radial fat-suppressed alternating-TR bSSFP sequence with T2-contrast for abdominal imaging during free-breathing. The sequence was tested in healthy volunteers and patients with pancreatic cancer and provided images of the abdomen during different respiratory motion states of diagnostic quality. This technique allows for visualization of pancreatic tumors during free-breathing, which is of great interest for radiotherapy treatment planning. Furthermore, the technique is extendable to high-resolution imaging in patients that are unable to perform long breathholds.