Introduction

Hyperpolarized (HP) ¹³C MR spectroscopic imaging (MRSI) enables quantitative monitoring of enzyme-catalyzed metabolism in human subjects. Using HP [1-¹³C]pyruvate, the rate of conversion of pyruvate to lactate (k_PL) via lactate dehydrogenase (LDH) can be computed per voxel in patients. However, ¹³C surface transmit/receive (T/R) coils do not have a homogeneous B₁ excitation profile, resulting in a gradient of decreasing flip angles for voxels increasingly farther away from the coil. Using the double angle method,¹ the coil’s excitation profile was measured and a contour plot mapping this gradient was computed and is shown in Figure 1 on top of a T₁-weighted phantom scan. The adjacent color bar indicates the scaling factor relative to the nominal flip angle. Tissue in the immediate proximity of the coil element is over-flipped by 200%. As the flip angle is provided as a parameter in the kinetic model used for estimating k_PL, the nominal flip angles require correction based on the coil’s B₁ excitation profile. The goal of this project was to develop and test a novel computational framework to improve k_PL accuracy using B₁ excitation correction in human studies.

Methods and Results

Data from three patient scans are summarized in Figure 2, with a representative case shown in Figure 3. The liver metastases were from the primary cancer origins listed in Figure 2. For each scan, we acquired an axial T₁-weighted spoiled gradient-echo (LAVA) anatomical reference, as shown in Figure 3(a), with the target lesion highlighted. In each scan, the ¹³C T/R coil was placed on the patient’s abdomen on the side closest to the target lesion to ensure coverage by the coil’s T/R field. ¹³C-pyruvate and ¹³C-lactate signals were acquired with a 2D echoplanar spectroscopic imaging (EPSI) sequence over 1 minute with a temporal resolution of 3 seconds and a spatial resolution of 1.2 cm in-plane. A constant flip angle scheme of 10° for pyruvate and 20° for lactate was implemented.

In MATLAB, the spectral data were quantified after zero- and first- order phase corrections. An SNR filter using pyruvate and lactate signals summed over time with an empirically-determined threshold was applied to remove noise voxels. Similarly, voxels outside of the coil’s sensitive region were excluded. The excitation profile of the ¹³C T/R coil was previously measured from a phantom scan using the double angle method.¹ Vitamin E capsules within the coil, which served as fiducial markers, were visible in the background T₁-weighted scan and are highlighted by yellow dots. Using the T₁-weighted references acquired during the patient scan, the previously measured excitation profile was rotated and translated in MATLAB to match the coil position during the patient scan as shown in Figure 3(b). As before, the adjacent color bar indicates the scaling factor relative to the nominal flip angle for each voxel. The resolution of the excitation profile was down-sampled to the coarser resolution of the ¹³C images using nearest-neighbor interpolation. This allowed for a voxel-by-voxel scaling of the nominal flip angles for pyruvate and lactate to reflect the excitation profile.

The k_PL values for each voxel were then computed using an inputless two-site model to generate k_PL maps.² The k_PL maps before and after the flip angle correction for an over-flipped tumor voxel are shown in Figure 3(c). The k_PL maps before and after the flip angle correction for an under-flipped tumor voxel are shown in Figure 3(d). The k_PL values for selected voxels are also summarized in Figure 2 along with the percent difference.

Discussion

The figure-8 coil provided adequate transmit power over the metastases despite the B₁ inhomogeneity. Monte Carlo simulations for the inputless two-site model demonstrated an inverse relationship between the relative B₁ error and the fractional k_PL error (defined as $$$\frac{k_{PL,fit}}{k_{PL,truth}} - 1$$$) as shown in Figure 4.² Positive errors in B₁ resulted in underestimations of k_PL and negative errors resulted in overestimations. In this study, over-flip led to an underestimation of k_PL (e.g. a -23.84% underestimation in Scan #3) whereas an under-flip led to overestimation (e.g. a +22.24% overestimation in Scan #3), which agree with prior simulations. The B₁ correction methods developed in this study can improve the quantitative accuracy of HP ¹³C metabolic MRI in cancer patients. As this method retroactively uses a phantom-derived B₁ excitation profile, in future studies, we plan to utilize an integrated fast B₁-mapping scheme for patient HP ¹³C-pyruvate MRI.³ This would allow for real-time power calibration that ensures the tissue of interest receives the prescribed flip angle. Nevertheless, due to the inherently inhomogeneous excitation profile of the surface coil, the method developed here would allow for improved k_PL estimations in regions outside of the tissue of interest which are over- and under-flipped or when there are multiple lesions of interest.

Acknowledgements

This work was supported by NIH grants U01 EB026412, R01 CA183071, and P41 EB013598.