Introduction

Dissolution dynamic nuclear polarization of ¹³C agents enables rapid measurement of cellular metabolism in vivo.¹ These measurements are challenging due in part to the transient nature of hyperpolarized (HP) signals, which persist for only around a minute in tissue and vary continuously due to the effects of perfusion, cellular uptake and metabolism. A typical HP experiment must therefore dynamically encode chemical shift information using specialized imaging techniques.

A variety of chemical shift encoding strategies have been used for imaging of metabolic agents by HP ¹³C MRI. Among the most widely used spectroscopic encoding methods for this purpose are spectrally selective RF pulses and multi-echo chemical shift decomposition (IDEAL). Specialized excitations, for example using compound spectral-spatial pulses, can perturb a single metabolite for selective imaging.² This approach is simple in practice, however the design of these pulses can be challenging depending on the spectral response needed and the capabilities of a given MRI system. Alternately, traditional broadband excitations can be used with spectroscopic readouts to obtain individual chemical shift images. Among spectroscopic readout strategies, IDEAL is appealing since it constrains encoding in a way that is optimized for the specific set of chemical shifts that are known to be present in a given HP experiment.³ In this work, we compare the SNR performance of selective excitations and IDEAL readouts for a typical preclinical imaging experiment using [1-¹³C]pyruvate.

Methods

We compared SNR performance of IDEAL and a spectral-spatial pulse designed for selective imaging of [1-¹³C]pyruvate and [1-¹³C]lactate at 7 T in numerical simulation and phantom experiments. In phantom experiments, IDEAL and spectral-spatial data were acquired for a range of effective excitation angles from a [¹³C]urea phantom using custom-built methods on a Bruker 7T MRI system equipped with BGA12-SHP gradients capable of delivering up to 660 mT/m and 4500 T/m/s. Numerical simulations were conducted in MATLAB to simulate data acquired with identical timing as phantom data, but with a range of effective excitation angles, T2* relaxation time constants and variances of complex zero-mean Gaussian noise in k-space data.

The 6.6 ms flyback spectral-spatial pulse considered in this work (Figure 1) was designed for individual excitation of pyruvate or lactate using a publicly available MATLAB toolbox.² An IDEAL encoding scheme using a series of eight 1.6 ms, 5 kHz SLR pulses with a TE shift of 0.52 ms was chosen for optimal signal averaging in the resulting chemical shift images (Figure 2). The same 2D singleshot flyback EPI imaging readout was used for both chemical shift encoding schemes (Figure 3). IDEAL was applied to raw k-space data with correction for chemical phase evolution along each imaging readout.⁴ No fieldmap demodulation was applied, and without this correction IDEAL is a linear transformation from the temporal to spectral domain with a magnitude and phase response shown in Figure 2. IDEAL data were acquired with minimum inter-echo delays to minimize T1 effects.

In order to facilitate comparison to individual spectral-spatial pulses, a variable excitation angle scheme was used for IDEAL across the eight pulses applied in order to provide a consistent level of transverse magnetization (M_Z) without consideration of relaxation effects. Individual excitation angles (θ) were calculated from the desired effective excitation angle (Θ) imparted by the series of N pulses as:⁵

$$\theta_i = \arcsin \left( \frac{\sin \Theta}{\sqrt{N} M_{Z,i-1}} \right) = \arcsin \left( \frac{\sin \Theta}{\sqrt{N} \prod^{i-1}_{j=0} \cos \theta_j} \right)$$

Results

The SNR performance of our single-band spectral-spatial RF pulse and our eight echo IDEAL encoding strategy are summarized in Figure 4. The SNR of images obtained by selective excitation and IDEAL encoding were found to be approximately identical in numerical simulations. Under no conditions were the mean SNR values of selective excitation and IDEAL images separated by greater than one standard deviation about their respective means. In thermal phantom experiments, IDEAL encoding consistently demonstrated a greater SNR than spectral-spatial excitation. Across all effective excitation angles measured, the mean ratio in mean SNR obtained by IDEAL relative to selective excitation was 32%. The greatest ratio in mean SNR measured for IDEAL relative to selective excitation was 42% at an effective excitation angle of 20 degrees.

Discussion

MRI scan parameters for HP experiments must be set carefully to provide optimal performance given the non-renewable nature of the imaging signal and the relatively low amplitude of signals produced via metabolic conversion in vivo. In this work we compared the SNR performance of two commonly used chemical shift encoding strategies for preclinical HP MRI experiments using [1-¹³C]pyruvate. A vital consideration in comparing these spectroscopic imaging strategies was the concept of effective flip angle, which captures the amount of longitudinal magnetization expunged to generate a set of images. While very similar SNR values were seen in numerical simulation, IDEAL demonstrated an SNR improvement of about 32% in thermal phantom experiments. Further study is needed to determine the reason for the greater SNR of IDEAL imaging in phantom experiments. Lastly, we note that the results presented are subject to the capabilities of the preclinical 7 T MRI system used in this work. Different results may be expected for MRI systems operating at different field strengths or with different gradient capabilities.

Acknowledgements

This work was supported in part by the National Institues of Health (R01-DK105346, R01-CA211150, P30-CA016672). The content represents the views of the authors and not of any sponsoring organization.