Frequency-selective hyperpolarized [1-¹³C]pyruvate EPI provides unparalleled acquisition speeds^[1], which are essential for volumetric, time-resolved metabolic mapping in vivo. In contrast to spectroscopic imaging techniques, echo-planar spatial encoding is notoriously sensitive to variations in the local resonance frequency, resulting in pixel shift artifacts along the blip encoding direction according to Eq. 1: y_shift(r) = y_true(r) + df(r) / BW_pp . Static B₀ field mapping methods^[2] represent the gold standard approach for correcting EPI distortion via Eq. 1; however, the true df(r) for the ¹³C signals is unavailable before the hyperpolarized injection, and running a field mapping acquisition wastes precious magnetization. Inhomogeneities induce distortion in the metabolic maps, compromising their spatial registration with respect to the underlying anatomy^[3]. To address this, we have developed a dual echo EPI sequence and non-iterative reconstruction that is capable of encoding the field map directly from the ¹³C signals. This ensures accurate spatial registration without any loss in SNR efficiency, and makes frequency-selective EPI more robust to B₀ inhomogeneities.

All gradient waveform design and image reconstruction was performed in Matlab (The MathWorks Inc., Massachusetts). The dual echo EPI sequence was implemented by removing every other blip trapezoid in a symmetric EPI readout such that each k_x line is traversed twice (figure 1). The result is two EPI images with opposing readout polarity, separated by an echo time difference of ΔTE = 0.892 ms (TR/TE = 56/26 ms, 128×16×12 matrix, 64×8×6 cm³ FOV, 5×5×5 mm³ nominal resolution). The phase of the complex quotient of the dual echo EPI images produces an estimate df(r)* in distorted coordinates. The estimate df(r)* is corrupted by an additional offset that arises from eddy currents, which cause odd/even echo asymmetry and erroneous phase accrual. To account for echo asymmetry, a fully phase encoded^[4] dual EPI reference scan was acquired on proton using the same G_x readout gradient amplitude as the ¹³C acquisition. G_y and G_z waveforms were scaled down by γ_¹³C / γ_¹H to provide 16×8×6 cm³¹H coverage. The residual between phase estimates computed from matching and opposite readout gradient polarity images was used for correcting df(r)*. A flowchart summarizing the signal processing pipeline is shown in figure 2. The final ¹³C phase map estimate in distorted EPI coordinates was used to unwarp the dual echo EPI images according to Eq. 1.

Imaging was performed on a GE MR750 3T MR scanner using a dual tuned T/R ¹H-¹³C rat coil (GE Healthcare, Waukesha, WI). Sprague Dawley (650g) rat images were obtained in accordance with a protocol approved by the institutional animal care and use committee. The net flip angle per volume for pyruvate and lactate was calibrated to 10° and 80°, respectively. Two consecutive imaging experiments were conducted on the same rat, ~25 mins apart. Imaging commenced as 3 mL of 80 mM pre-polarized [1-¹³C]pyruvate solution (SpinLab DNP Polarizer, GE) was injected over 12 s via tail vein catheter. An estimate of the true in vivo centre frequency was used for the first experiment, while a deviation of -50 Hz was used to induce spatial mis-registration in the second experiment.

Reconstructed metabolic maps were 3X spline interpolated and overlaid on corresponding axial T2-FSE images. Data acquired during the 2^nd experiment (-50 Hz off resonance) is shown in figure 3. Corrected images exhibit superior alignment of the signals arising from the aorta and kidneys with respect to the underlying anatomy. Time courses of the total pyruvate and lactate signals within a manually traced ROI comprising both right and left kidneys illustrate the difference between registered and mis-registered metabolic maps.

The uncorrected lactate time course reveals signal contamination from pyruvate-hydrate within the kidney. For displaying the lactate images, only time points from t = 30-60 s were summed to minimize the appearance of the pyruvate-hydrate contamination. Spectral contamination is unavoidable with large errors in frequency setting, and cannot be resolved with the proposed dual echo EPI sequence; however, the phase map estimates may be useful as a quality metric when interpreting lactate images.

We have demonstrated a novel acquisition and reconstruction scheme for providing tolerance to erroneous frequency variations including B₀ inhomogeneities and eddy currents in frequency-selective ¹³C EPI of hyperpolarized compounds. The method is non-iterative, simple to implement, and the degree of apodization is the sole tuning parameter. This parameter was manually adjusted to generate the images in figure 3, and its selection represents a tradeoff between ¹³C phase map SNR and spatial accuracy. The automated selection of this parameter will be investigated in future work.