Coil profile estimation strategies for parallel imaging with hyperpolarized ¹³C MRI

Abstract

Purpose To investigate auto‐ and pre‐calibration coil profile estimation for parallel imaging reconstruction of hyperpolarized 13C MRI volumetric data. Methods Parallel imaging reconstruction was studied with 3 different approaches for coil profile estimation: auto‐calibration, phantom calibration, and theoretic calibration. Acquisition was performed with a 3D stack‐of‐spirals sequence with spectral–spatial excitation and Cartesian undersampling. Parallel imaging reconstructions were done with conjugate gradient SENSE and 3D gridding with inhomogeneity correction. The approaches were compared in simulations with different SNR, through phantom experiments, and in an in vivo pig study focused on the kidneys. All imaging was done with a rigid home‐built 12‐channel 13C receive coil at 3T. Results The phantom calibrated and theoretic approaches resulted in the best structural similarities in simulations and demonstrated higher image quality in the phantom experiments compared to the auto‐calibrated approach. In vivo mapping of pyruvate uptake and lactate conversion improved for accelerated acquisitions because of a better temporal resolution. From a practical and image quality point of view, use of theoretic coil profiles led to improved results compared to the other approaches. Conclusion The success of the theoretic coil profile estimation demonstrates a negligible effect of load on sensitivity profiles at the carbon frequency at 3T. Through theoretic or phantom calibrated parallel imaging, accelerated 3D volumes could be reconstructed with sufficient sensitivity, temporal, and spatial resolution to map the metabolism of kidneys exemplifying abdominal organs. This approach overcomes a critical step in the clinical translation of parallel imaging in hyperpolarized 13C MR.

Three‐dimensional accelerated acquisition for hyperpolarized 13C MR with blipped stack‐of‐spirals and conjugate‐gradient SENSE

“…The duration of the combined excitation and readout was 70 ms 30 × 30 × 16 cm 3 reconstructed FOV, at a 5 × 5 × 10 mm 3 nominal resolution, achieved in 4 excitations with a spiral readout duration of 45 ms. 23 All data were acquired on an MR750 3T scanner (GE Healthcare, Waukesha, WI) with a locally built, fixed-geometry coil setup consisting of a linearly polarized 13 C volume coil for transmission and a rigid 12-element array for reception. 24 The 12 receive elements (with 130 mm diameter) were distributed in 2 rows mounted over a cylinder with a diameter of 300 mm. The receive array and coil setup are shown in Supporting Information Figure S1.…”

“…It has been demonstrated that loading effects have minimal effect on coil-sensitivity profiles at the carbon frequency at 3 T. 24 For fixed coil geometry, this enables the use of precalibrated sensitivity profiles in parallel imaging reconstruction of undersampled 13 C data. In this study, precalibration was based on simulated complex sensitivity profiles of the 12-element receive array by means of CST (Darmstad, Germany) RF modeling.…”

“…24 For all phantom and in vivo experiments, the G z gradient of the SPSP pulse was designed to limit excitation outside the encoded FOV and thereby minimize wraparound effects at the cost of nonuniform excitation efficiency within the FOV. 24 To compensate for the irregular profile, an inverse filter based on the simulated excitation profile was used in the reconstructions. The simulated excitation profile of the SPSP pulse is shown in Supporting Information Figure S3.…”

“…The apparent exchange rate from pyruvate to lactate was set to k PL = 0.025 s −1 based on previous experiments. 24 The flip angle was chosen based on maximum accumulated signal. With multiple excitations per volume, the effective flip angle is defined in relation to the resulting volume excitation after the series of smaller flip angles.…”

“…Both fully sampled and undersampled acquisitions were reconstructed using the same pipeline initially validated by simulated data in previous work. 24 The CG-SENSE reconstruction (i.e., iterative SENSE reconstruction by means of the CG algorithm 31 ) was performed based on source code given by "gpuNUFFT -Open-Source GPU Library for 3D Gridding" on GitHub. 32,33 The reconstruction requires 5 main inputs: (1) the 3D k-space data, (2) a 3D NUFFT forward operator to perform 3D gridding, (3) a 3D array of coil sensitivities, (4) the number of CG iterations, and (5) a regularization parameter for Tikhonov regularization.…”