Today, routine knee examinations consist of several 2D turbo spin echo (TSE) scans, each covering multiple slices in different orientation. These scans differ in contrast, usually including proton density (PD) and T2 weighting, both without and with fat suppression. A 3D TSE scan with high, isotropic resolution promises to replace typically two of these scans with similar contrast but different orientation, because images can be reformatted retrospectively to any orientation without loss of resolution. At the same time, a chemical shift encoding-based water-fat, or Dixon, TSE scan allows substituting two of these scans with similar basic contrast and same orientation, but without and with fat suppression, respectively. Consequently, two 3D Dixon TSE scans may suffice for routine knee examinations, one with PD and one with T2 weighting. However, the scan time of such 3D Dixon TSE scans is usually still too long for clinical practice, not to mention the increased risk of motion artifacts. Therefore, the purpose of this work was to explore the basic feasibility of exploiting the additional acceleration provided by the combination of compressed sensing (CS) and parallel imaging compared with parallel imaging alone for reducing the scan time of 3D Dixon TSE scans to that of conventional 3D TSE scans with fat suppression.We limited this preliminary investigation to the representative case of PD weighting. For reference, we took a conventional 3D PD TSE scan, once without and once with fat suppression. The 3D PD TSE scan without fat suppression had the following parameters: TSE factor: 60, echo spacing: 4.5 ms, shot duration: 300 ms, profile order: low-high radial, apparent TE: 27 ms, TR: 1000 ms, pixel bandwidth: 595 Hz. Using a SENSE acceleration of 5.0 (2.0 x 2.5), this scan took 3:18 min. The 3D PD TSE scan with fat suppression employed SPAIR and the following parameters: TSE factor: 63, echo spacing: 6.4 ms, shot duration: 438 ms, profile order: linear, apparent TE: 59 ms, TR: 1200 ms, pixel bandwidth: 345 Hz. Using a SENSE acceleration of 4.4 (2.0 x 2.2), this scan took 4:51 min. To introduce chemical shift encoding, the 3D PD TSE scan without fat suppression served as starting point. Using a multi-repetition strategy, the number of shots was doubled, and the readout gradient and acquisition window were shifted in every second shot, requiring in total at least twice the scan time. This was compensated by complementing SENSE with variable density Poisson disk sampling and an L₁ norm-based CS reconstruction using pre-calibrated coil sensitivities^1,2. In this way, two single-echo images were reconstructed separately, before performing the water-fat separation³. Compared to the 3D PD TSE scan without fat suppression, the 3D PD Dixon TSE scan had the following parameters: TSE echo spacing: 6.7 ms, shot duration: 451 ms, apparent TE: 41 ms, pixel bandwidth: 633 Hz. Using an effective CS-SENSE acceleration of 6.8 (2.6 x 2.6), this scan took 4:49 min, about as long as the 3D PD TSE scan with fat suppression.
Healthy subjects were examined on a 3T Ingenia scanner (Philips Healthcare, Best, Netherlands) using an 8-element knee receive coil (InVivo, Gainesville, USA). All scans covered a single knee with a FOV of 145 x 160 x 160 mm³ and an isotropic resolution of 0.7 mm.Representative results obtained in one of the subjects are summarized in Figs. 1-3. Images acquired for reference with the conventional 3D PD TSE scans are provided in Fig. 1. Corresponding images reconstructed from the described 3D PD Dixon TSE scan are shown in Fig. 2. They were produced by first generating the water and fat images displayed in Fig. 3 and then calculating linear combinations of them to obtain contrasts similar to those in Fig. 1.The results obtained so far suggest that the addition of compressed sensing may indeed allow the acquisition of corresponding images without and with fat suppression in similar scan times as the acquisition of one of these images today. The examples shown in Figs. 1-3 indicate that both SENSE, at least in the scan without fat suppression, as well as CS-SENSE were operated close to the limits of the employed coil array and the sparsity of the underlying data. Further acceleration, or artifact reduction, may be expected from a multi-echo instead of a multi-repetition approach to chemical shift encoding⁴.