Recently, dedicated knee coils have become available for clinical 7.0T imaging. For the latter, reliable fat suppression is mandatory which is, however, challenging due to increased magnetic field inhomogeneity, inhomogeneous RF transmit field and higher specific absorption rate (SAR). Standard techniques such as spectral pre-saturation with inversion recovery (SPIR) or spectral adiabatic inversion recovery (SPAIR) increase SAR and lengthen the acquisition time. An alternative is slice selective gradient reversal (SSGR) which uses gradients of opposing polarity for the excitation and refocusing pulse (1). SSGR is expected to perform well at 7.0T without introducing additional scan time or increasing SAR (2,3). Purpose of this study was to evaluate different fat suppression techniques for clinical 7.0T knee MRI.Local ethics board approved study with written informed consent from eight healthy volunteers (mean age 31±4 years, six males). Volunteers were imaged at 7.0T (Achieva, Philips Healthcare, Cleveland, OH) using a dedicated 28-channel TX-knee coil (QED, Quality Electrodynamics, Mayfield Village, OH) and at 3.0T (Skyra, Siemens Healthcare, Erlangen, Germany) using a 15-channel TX-knee coil (QED). At 3.0T, an axial proton-density-weighted turbo spin echo (PDw-TSE) sequence with SPAIR from the hospital`s clinical standard protocol, and at 7.0T, a series of five PDw-TSE sequences were acquired: a) without fat suppression, b) with SPIR, c) with SPAIR, d) with SSGR and e) with the combination of SSGR+SPIR. Except for the fat suppression technique, all other imaging parameters were kept identical for all acquisitions (field-of-view: 160x160mm², TR: 3800ms, TE: 35ms, voxel size: 0.35x0.45x2.50 mm³, 15 slices, 1 signal average, bandwidth: 216 Hz/pixel, acquisition time: 3:40 min). Each 7.0T sequence contained an additional noise scan without gradients and RF to allow pixel-wise calculation of signal-to-noise ratio (SNR) maps (4). SNR values were extracted from regions-of-interest (ROIs) within the cartilage, bone marrow, joint fluid, muscle and fat. Contrast-to-noise ratio (CNR) values were calculated between these tissue types. All MR measurements were repeated on a dedicated water-fat phantom (ultrasonic gel and swine fat). Additionally, the image quality and fat suppression was analyzed by two independent radiologists using 5-point Likert scales. Kappa statistics were calculated to evaluate inter-reader agreement. Quantitative and qualitative results from the different sequences were compared using paired sample t-tests and Wilcoxon signed rank tests. A p-value of <0.05 was considered statistically significant.

Quantitative analysis: In the phantom experiment, fat was only partially suppressed using SPIR and SPAIR. SSGR and SSGR+SPIR suppressed the fat signal much more compared to the original fat signal without any suppression (below 15% of the original signal) (Fig. 1). Figure 2 shows examples of acquired images in vivo at both field strengths and corresponding SNR maps. At 7.0T, SSGR and SSGR+SPIR yielded significantly lower SNR in fat compared to SPIR and SPAIR (p<0.001, Fig. 3). SSGR and SSGR+SPIR yielded significantly lower SNR in cartilage compared to SPIR (p≤0.002) but similar SNR compared to SPAIR (SSGR: p=0.105, SSGR+SPIR: p=0.132). Relative SNRs of fat demonstrated that the SPIR technique reduced the fat signal to 48±6%, SPAIR: 22±1.6%, SSGR: 12±0.9% and SSGR+SPIR: 10±0.25% (Fig. 1 and 3). Evaluation of CNR showed superior contrast between muscle-fat, cartilage-fat, fluid-fat, and fluid-cartilage for the SSGR method (Fig. 4).

Qualitative analysis: In-vivo images showed improved image quality for all 7.0T methods compared to the clinical reference 3.0T SPAIR image (all p≤0.001) (Tbl. 1c,d). Compared to 3.0T, the grade of fat suppression was rated lower for SPIR (mean±SD, 2.0±0.00) and SPAIR (3.0±0.00) but higher for SSGR (4.0±0.25) and SSGR+SPIR (5.0±0.00). The homogeneity of the fat suppression was rated lower for 7.0T SPIR (3.69±0.48) compared to the clinical reference 3.0T SPAIR image (3.94±0.44). The homogeneity was better than the clinical reference with 7.0T SPAIR (4.0±0.0), SSGR (4.19±0.40) and 7.0T SSGR+SPIR (4.31±0.48). However, only for SSGR+SPIR the homogeneity improvement over the clinical reference image was finally statistically significant (Tbl. 1c,d). The SSGR method provided a strong, stable and homogenous fat suppression for clinical knee imaging at 7.0T. Quantitatively no significant differences in the grade of fat suppression was found between SSGR and SSGR+SPIR, but SSGR+SPIR yielded qualitatively better images in terms of grade of fat suppression and homogeneity of fat suppression than SSGR as assessed by two independent radiologist. Compared to standard spectrally-selective suppression methods at 3.0T, the SSGR technique can provide stand-alone or in combination with SPIR fat suppression in better quality, a shorter scan time and without a SAR elevation.At 7.0T, fat saturation for clinical knee imaging using SSGR and the combination of SSGR and SPIR was superior compared to methods based on spectrally selective RF pulses.