INTRODUCTION

APT/CEST-MRI has shown promising applications in neuro-oncology and translation to body-oncology is ongoing¹, however, robustness is compromised with fat-signals as main error source². To extract CEST effects (1-5%), fat-correction^3,4 needs to be precise to <1%. For APT/CEST-MRI, combined RF-saturation and SPIR fat-suppression (selective π-pulse) increases signal intensity near the fat-frequency (relative to non-saturated S₀, Fig.2d). With the fat-signal saturated near -3.3ppm, the SPIR inversion pulse has no effect, and fat-signal reappears via saturation recovery. We observed that the signal near the fat frequency closely follows the mirrored line shape of a multi-peak fat model and investigate, if this line shape in combination with direct water saturation (DWS) and MT could be favorably used for background fitting.

METHODS

Phantom tests were performed at 3T with dual-channel RF-transmission (Ingenia, Philips, NL) using a 16-channel head-coil and 2^nd-order B₀-shimming. Two phantoms were used with sunflower-oil on top and (1) phantom fluid (water+0.8g/l CuSO₄,T₁=450ms, no APT-effect) or (2) egg-white (APT-effect), coagulated (10min at 60°C) to obtain MT-effects. An oblique slice provided a range of fat contents (Fig.1).
For saturation, 40×50ms pulses were alternatingly transmitted via the RF-channels (100% duty-cycle), with T_sat=2s and B_1,rms=2μT. A single-shot 2D fast-spin echo sequence with/without SPIR fat-suppression was used with FOV=(160mm)², voxel size 1.2×1.2×8mm³, TR/TE=5700/7.0ms, FA=90°, 120° refocusing, centric k-space-ordering, and (i) N=42 equidistant frequency offsets, 0.4ppm steps, range -8..8ppm, acquisition time T_acq=4min, or (ii) N=17 offsets: {-1560(S₀),±7.8,±6.3,±5.1,±3.9,±2.9,±1.8,±0.9,±0.3}ppm, T_acq=2min. A 3-point multi-acquisition GRE-Dixon sequence, same geometry and TE, was used for B₀ mapping: TR=14 ms, ΔTE=0.6 ms, FA=35°, 8 averages, T_acq=45s. Z-spectra were interpolated to (i)126/(ii)100 points including B₀-correction. Two fitting models including Lorentzian (DWS) and Gaussian (MT) components were used with/without SPIR fat-suppression, with “-” or “+” sign for the fat contribution in Eq.1, respectively.
$$S_{model}(f)=S_0-\mid A_L \frac{w_L^2}{f^2+w_L^2}+A_G\frac{1}{\sqrt{2\pi}w_G}\ e^{\large{-\frac{1}{2}(\frac{f}{w_G})^2}} \pm A_F\ \sum_{l=1}^{N_l}\frac{w_F^2}{(f-f_l)^2+w_F^2}\ e^{-2\pi i f_l \ T\small{E}} \mid \ \ \ [1]$$
5 parameters were fitted: A_L, A_G, A_F (amplitudes) and w_L, w_G (Lorentzian/Gaussian width). Fixed fat linewidths w_F=110 Hz and fat frequencies f_l (N_l=7 lines) were used, similar to common multi-peak Dixon water-fat separation. After B₀-correction, centered distributions can be assumed. 11 interpolated points were selected for fitting: {±7.0,±4.5,-3.4,-2.3, ±1.3,-0.9,±0.3}ppm, explicitly excluding the range of amide (or amine) CEST effects [+1.3…+4.5ppm]. APT/CEST signals are extracted as APT#(+3.5ppm) as residuals from the background fit, similar to extrapolated MT reference (EMR) fitting⁵. NOE#(-3.5ppm) serves as quality check for the fit. Fitting was performed for each voxel with a custom-made ImageJ (imagej.nih.gov) plugin.

RESULTS

Fig.2 shows example Z-spectra, demonstrating that the model closely fits for various fat content (a/b/d/e), MT-background and APT-effect (c/f) without (a/b/c) and with SPIR fat-suppression (d/e/f). Similar data quality was obtained with protocol (ii) (N=17, data not shown). Without fat-suppression, the spectral shape is strongly altered depending on fat contents but well captured by the model (Fig.2a/b, in-phase TE). The APT-effect from egg-white proteins is visible as deviation between [+4…+2]ppm, excluded for fitting (2c). The MT-effect appears as broad attenuation, successfully modeled by the Gaussian component. Z-spectra with fat-suppression are precisely fitted including the signal overshoot near -3.3ppm (2d/e). Here, DWS amplitude/symmetry is hardly changed by fat effects. MT-effects are captured by the fit, while the APT signal is apparent as residual around 3.5ppm (2f). Some slight deviations are observed around -5 and -8ppm.
Fig. 3 shows processed APT# (3a/d/g), NOE# (3b/e/h) and MTR_asym (3c/f/i) images for phantom1 (3a-f) and phantom2 (3g-i), respectively, in a range of ±5%. While MTR_asym shows negative contrast for fat without SPIR (3c/i), the signal overshoot leads to false hyperintensity (red) with SPIR (3f). APT# results show homogeneously low (green) values (3a). With SPIR, high fat content areas show some dropouts (3d). APT# is hyperintense in egg-white containing areas (3g). NOE# images show overall low values, except for some dropouts/hyperintensity for SPIR and high fat (3e).

DISCUSSION

The results largely confirm the experimental validity of the chosen fitting model with homogeneously low fit residuals for APT#/NOE# over a large range of fat fractions. To our knowledge it is the first use of a mirrored shape of the fat spectrum with fat-suppression. APT-effects can be obtained as residuals around 3.5ppm (APT#). Dropouts/hyperintensities with SPIR arise from low signal levels in high fat areas after fat-suppression impeding reliable fits. However, areas with large fat fractions (>90%) contain little CEST information and may be masked. Deviations at larger negative offsets (-5…-8ppm) indicate that fitting points >8ppm could be included to better separate fat from MT. Fat-saturation influences spectral lines differently but were negligible for in-phase TE. The fit model includes fat phase relations for different TE, but Z-spectra may be difficult to fit². However, sequences with fat-suppression show smaller fat-amplitudes and no reduction of the DWS peak which may facilitate stable fitting, to be investigated further. While limited to 2D here, minimization of the acquired Z-spectral points and the addition of parallel imaging would enable 3D protocols in clinically acceptable times.

CONCLUSION

We have proposed fitting models and protocols with/without SPIR to obtain fat-signal corrected APT# signals at various fat content and MT background levels. Z-spectral results from water/fat/APT/MT phantoms are promising. In vivo breast studies are underway to explore the applicability in body-oncology.

Acknowledgements

Support by NIH grant 1R01CA252281-01A1.