Prior human HP ¹³C-pyruvate MR studies used older acquisition methods which provided either time-averaged 15s 3D MR spectroscopic imaging data or dynamic data that was limited to 1D spectroscopy or a single 2D slice, and did not cover the entire prostate^1,2. In order to meet the clinical need of full gland coverage to detect all regions of cancer, a 3D dynamic compressed-sensing (CS) EPSI sequence was developed and tested first in phantoms and transgenic models of prostate cancer (TRAMP) before initiating human studies. Clinical studies require larger imaging volumes and reduced peak RF power for the large RF transmit coil. This study was designed to translate prior acquisition methods to enable human 3D dynamic HP acquisitions by addressing these challenges through the optimization of pulse design and sequence parameters.

Sequences: Reduced peak power spectral-spatial excitation pulses for a clinical transmit coil were designed using the SS-RF toolbox developed by Larson et al.³ These pulses incorporated multiband RF excitation and a variable flip angle schedule aimed to maximize total lactate signal. Our pre-clinical 3D CS-EPSI sequence used double spin-echo refocusing (“DSE” sequence)⁴, but these high-flip pulses can result in substantial magnetization losses with inhomogeneous B1 fields⁵. For clinical studies, we removed these refocusing pulses (“FID” sequence), and used a modified CS reconstruction including phasing and enforcing spectral sparsity to maintain narrow linewidths. Additional sequence parameters include TR = 150ms, TE = 6.3ms, spatial resolution: 3.3mm x 3.3mm x 5.4mm (mice), 6.7mm x 6.7mm x 10.8mm (rats), or 8mm x 8mm x 8mm with FOV=9.6x9.6x12.8cm (humans), temporal resolution = 2s, acquisition window = 36s.

MRI experiments: Figure 1 shows the sequence diagram and RF pulse design. To test the sequence in vivo prior to human studies, 11 sets of imaging studies were conducted on 6 transgenic mice of prostate tumor (TRAMP) and 3 healthy Sprague Dawley rats. [1-¹³C]pyruvate and ¹³C urea were co-polarized using a GE SpinLab polarizer for 2 hours, yielding 20-30% polarization. For the TRAMP studies, a 350ul bolus was injected over 15s, while rats received a 3ml bolus over 12s, both administered via a tail vein catheter. The sequence was initiated at t=15 seconds after injection began. In the clinical setup, a clamshell coil was used for ¹³C transmit, and a dual-tuned endorectal coil for reception¹.

Image Reconstruction and Dynamic Modeling: Quantitative estimation of metabolism and perfusion were performed by calculating the area-under-curve (AUC) for lactate and pyruvate, then applying the compartmental exchange models to the HP-¹³C dynamic curves. Multiband excitation, variable flip angles and T1 relaxation were taken into account for signal-versus-concentration correction⁶.

In preclinical studies, we observed similar performance in the new “FID” sequence compared to the previous “DSE sequence”. The mean SNR was comparable between our “DSE” data (33.47±6.96 dB) and the 8 “FID” acquisitions (36.28±3.98dB) on TRAMP tumor (P>0.3). In the rat scans, both FOV and voxel size were doubled from mice, where high SNR was found in both rat kidney (43.84±4.68dB) and liver (37.75±4.75dB), showing robustness to scaling. While the “FID” sequence suffers from the inherent T2* loss, the 180° refocusing in the “DSE” sequence may spoil the HP magnetization at coil edge.

The spectral-spatial excitation pulse for clinical studies yielded a peak B1 of 0.597G, a 67% reduction compared to our previous preclinical designs (1.796G), and phantom data showed good agreement with the simulated performance (Fig.1). Figure 2 demonstrates full-gland coverage in a patient with bilateral biopsy-confirmed prostate cancer following the injection of sterile GMP [1-¹³C]pyruvate polarized to 37% in the 5T SpinLab. Figure 3 shows the dynamic pyruvate and lactate images acquired with a time resolution of 2s and a spatial resolution of 0.5cm³. This 5 dimensional (3 spatial, 1 spectral, 1 temporal) data (Fig.4) enable the detection of the dynamics of the conversion of pyruvate to lactate catalyzed by LDH that is known to be greatly up-regulated in prostate cancers.

This new clinical 3D dynamic acquisition method incorporating a new spectral-spatial RF, “FID” readout and modified CS reconstruction addressed the challenges of larger imaging volumes, reduced available peak RF power and increased B1 inhomogeneity required for human studies. The 3D dynamic CS-EPSI exhibited excellent performance and robustness in phantom, preclinical, and patient exams.