MR physicists and clinicians interested in spine imagingSpine MRI exams, be it for looking at cord pathologies, infection or investigating lower back pain, typically consist of multiple 2D acquisitions with different contrast weightings [1]. Techniques that extract proton density (PD) and T1, T2 relaxation maps simultaneously from a single acquisition and use the multi-parametric maps to generate synthetic images of any desired image contrast have been demonstrated in recent years, mostly in the brain [2-7]. Such an approach could potentially shorten spine MRI exams if for e.x. T1, T2 and PD images could be synthetically generated and additionally quantitative T2 maps could be obtained from a single scan in a lower back exam. But even routine day-to-day spine MRI is fraught with technical challenges-motion from swallowing (cervical spine), cerebrospinal fluid (CSF) flow, respiration, B0 inhomogeneity, B1 non-uniformity all pose impediments to consistent image quality [8]. So we investigated if it would be feasible to derive synthetic images using a 2D quantitative mapping technique, MAGiC, in the 2 most often scanned spine regions-lumbar and the cervical spine.

In MAGiC, a saturation delay multi echo fast spin echo (FSE) sequence is used to acquire multiple saturation delay times and multiple spin echo times [9]. T1 and T2 fitting to the different delay and echo times, computation of PD from the scaling of the curves and further processing is performed by the SyMRI processing pipeline (SyntheticMR, Linköping, Sweden).

1 volunteer each was scanned at the Lspine and Cspine with informed consent in accordance with IRB guidelines of the site on a 3T scanner (GE MR 750 Waukesha, WI) using six and nine elements of a spine array for the lumbar and cervical regions respectively. Data from four saturation delay times and 2 echo-times was collected with the above mentioned FSE sequence (FOV= 26 cm (Cspine: 18 cm), phase FOV factor=0.8, slice thickness/gap=4/1 mm (Cspine: 3/1 mm), 20 slices, TR/TE1/TE2 = 4000/21.5/85.5 ms, echo train length=12, matrix: 320x256, 20 slices, scan time=9:30 mins) with phase encoding along the superior/inferior (S/I) direction, spatial saturation bands placed superiorly and inferiorly to the phase FOV and with full sampling to retrospectively explore optimal undersampling strategy,

Two reconstruction approaches were explored on the fully sampled data: 1) k-t adaptive ARC (kat-ARC) [10] with simulated staggered time shifted sampling pattern across the temporal phases, and 2) ESPIRit, an iterative autocalibrated parallel imaging method that uses eigenvector maps [11] with simulated random undersampling pattern. Additionally, a local low-rank (LLR) constraint was added to exploit data redundancy along the parameter dimension [12]. Reconstructed images were post-processed by the SyMRI pipeline.

The spine array used in this work offered limited acceleration capabilities only in the S/I direction. Addition of local low rank constraint to ESPIRiT provided appreciable improvement in SNR and image quality, even for the small size of the parameter dimension. An example in the Lspine from the first saturation delay timepoint is shown in Figure 1. kat-ARC and ESPIRiT-LLR reconstructions provided comparable image quality at two-fold acceleration, while it was possible to achieve slightly higher acceleration of up to 3 folds with the second approach albeit with some SNR penalty. T1 FLAIR, T2 weighted images and R2 maps of the of the Lspine processed from images reconstructed by ESPIRiT-LLR and T1, T2 weighted images and R1 map of the Cspine processed from images reconstructed by kat-ARC, from simulated 2X acquisitions of 4 minutes and 45 seconds, which is slightly shorter than the time typically taken to acquire 2 separate T1w and T2w sequences, are shown in Figures 2 and 3 respectively. The Cspine images were noisier because of being acquired at a higher resolution than the Lspine and had some flow ghosting. In this work we showed the feasibility of deriving synthetic MR images in the spine based on a 2D quantitative mapping technique, with some challenges in the Cspine where more robust motion compensation is needed. Additional derived contrast weightings such as phase sensitive inversion recovery (PSIR) or double inversion recovery (DIR) could potentially add value by helping visualize cord pathologies. While a 3D quantitative parameter mapping method would allow for multi-planar reformats, 2D acquisition might be more suited to the spine due to motion considerations. Acceleration and exploitation of data redundancy in the parameter dimension, and coil arrays with better parallel imaging capabilities are crucial to achieving SNR-efficiency in such parameter mapping techniques.