Introduction

The characterization of tissue oxygenation (pO₂) with MRI, known as MR-oximetry, is interesting for the study of diabetes and fatty liver disease¹, and in cancer, where the hypoxia commonly observed in solid tumors is known to reduce the efficacy of radiotherapy². Fat R₁ is a promising biomarker for MR-oximetry, more sensitive to oxygen variations (∆O₂) than global R₁ because oxygen is more soluble in fat than in water³. Existing techniques to measure fat R₁ require an important presence of fat³, are limited to 2D imaging with long scan times¹, or have only been demonstrated in single-voxel^4,5 or region of interest (ROI) measurements⁶. To address these limitations, we present Fat DESPOT, a volumetric technique to map fat R₁ on a voxel-wise basis using a variable flip angle (VFA) approach.

Methods

Dairy cream (fat fraction f = 35%) at room temperature was used as an approximation of fat in human tissue, in a phantom with six 50-ml cream-filled vials with controlled variations of oxygen concentration. Oxygenation was set by bubbling the cream with nitrogen (N₂) or oxygen (O₂) for controlled periods of time. After imaging, the O₂ concentration was measured in each vial with a dissolved O₂ meter (Orion Star A323, Thermo Fisher Scientific).

Data were collected in a 3 T scanner (Ingenia, Philips) with an extremity coil. For Fat DESPOT R₁ mapping, 3D three-echo gradient echo (MGRE) data were acquired with four flip angles (FA). For each FA, the sequence was run twice while changing the first echo time (TE), resulting in six echoes with an effective ∆TE of 1.2 ms. B₁ mapping was performed using the dual angle method⁷. Reference R₁ was measured in each phantom vial using a single-voxel STEAM MRS sequence with a series of mixing times (TM). The fat spectrum of dairy cream was obtained from STEAM data with minimum TM. Sequences parameters are listed in Table 1.

For voxel-wise fat and water R₁ mapping with Fat DESPOT, the MGRE magnitude data was fitted with the following equation (with five fitting parameters: S₀, R₂^*, f, R_1w, R_1f). This two-compartment model, based on previous work⁶, includes the measured fat spectrum (relative amplitude α_p and frequency shift relative to water Δω_p) and a B₁ correction (on θ_n). The fit was performed separately with four (3°, 6°, 15°, 34°) and two (6° and 34°) FAs to evaluate the possibility of reducing the scan time. To assess the variability of Fat DESPOT, the MGRE measurements with two FAs were repeated seven times in one vial.

$$S(TE_i,\theta_n,TR)=S_0e^{-R_2^*TE_i}\left[(1-f)\frac{(1-e^{-R_{1_w}TR})\sin(\theta_n)}{1-e^{-R_{1_w}TR}\cos(\theta_n)} +f\frac{(1-e^{-R_{1_f}TR})\sin(\theta_n)}{1-e^{-R_{1_f}TR}\cos(\theta_n)}\sum_{p=1}^x(\alpha_pe^{i∆\omega_pTE_i})\right]$$

The STEAM MRS spectra were fitted with a nine-peak spectral model (eight fat peaks and the water peak) using least-squares fitting, after zeroth-order phase correction and apodization. For R₁ calculation, the area (S) of three peaks (water and the two most abundant fat peaks, methylene and methyl) was fitted as a function of TM⁸ with non-linear least-squares using:

$$S=Ae^{-R_1TM} $$

Linear fits were performed on the estimates of R₁ as a function of the measured O₂ concentration, for Fat DESPOT and MRS STEAM, to extract the O₂ relaxivity (slope in ms^-1/ %O₂). Fat and water R₁ from Fat DESPOT were extracted in circular ROI in each vial. The experiment (Fat DESPOT and STEAM MRS) was repeated three times to evaluate the replicability.

Results and discussion

High quality Fat DESPOT fits were obtained in all voxels (R²>0.95). Fat R₁ showed a higher O₂ relaxivity than water R₁ (Figure 1) and a stronger linear correlation (Table 2), as anticipated¹. The four- and two-FA fits led to the same conclusions, enabling the reduction of scan time by a factor of two. Fat DESPOT also showed a low variability. The coefficients of variation (CV) over the seven scan repetitions of the mean fat R₁, water R₁, and fat fraction were below 3.5 %, much lower than a previous report using a similar method with a CV reaching 25%⁶.

With STEAM MRS, the methyl and methylene peaks were well resolved (Figure 2). These peaks showed a significant linear correlation between R₁ and O₂ concentration (Table 2), consistent with a previous report⁴. The methyl and methylene R₁ were more sensitive to O₂ variations than the water R₁ (Figure 3).

The O₂ dependence in Fat DESPOT was consistent with the reference technique. Both showed the expected linear relationship between R₁ and oxygen concentration, with a greater slope for fat R₁ than water R₁. The O₂ relaxivity for fat (Table 2) was comparable between the two techniques. The results of Fat DESPOT and STEAM MRS were consistent between the three experiments.

Conclusion

This work demonstrated the feasibility of using a 3D VFA approach to measure fat R₁ at 3 T in the context of oximetry. Fat DESPOT provides results consistent with literature and MRS measurements, and addresses the main limitations of existing methods through high-resolution 3D coverage in a clinically acceptable measurement time (<15 min). This technique could provide relevant information for personalized radiotherapy treatments in oncology to overcome the effects of hypoxia. It could also provide valuable information in other contexts where hypoxia is of interest, such as obesity-related metabolic diseases.

Acknowledgements

The authors acknowledge the Body Magnetic Resonance research group (Technical University of Munich) for sharing their MRS processing software, the developers of the ISMRM fat water toolbox (http://www.ismrm.org/workshops/FatWater12/ data.htm), and Guillaume Gilbert (Philips Healthcare, Inc.), Atiyah Yahya (University of Alberta), and Jamie Near, Norma Ybarra, Zaki Ahmed, and Stella Xing (McGill University) for useful discussion. This work was funded by the Montreal General Hospital Foundation, the Research Institute of the McGill University Health Centre, and a Discovery Grant from the Natural Science and Engineering Research Council of Canada (NSERC). VF acknowledges fellowship support from NSERC, and partial support from the NSERC CREATE Medical Physics Research Training Network (Grant number: 432290).