The spinal cord (SC) is affected in demyelinating diseases of the central nervous system, such as Multiple Sclerosis [1]. Using quantitative Magnetisation Transfer (qMT) methods, it is possible to extract measures related to macromolecular tissue content, which have been shown to be specific for myelin in the brain and SC [2].

However, performing qMT in the SC is challenging, mostly due to the prohibitive scan times required to acquire multiple high resolution images in order to accurately estimate all model parameters. Therefore qMT has not found widespread application in vivo in the SC, with just a single study previously published [3]. Efforts have rather aimed to develop simplified versions of the rigorous qMT [4], or optimise semi-quantitative approaches [5,6].

We explored the possibility of performing qMT in vivo in the SC within a clinically feasible acquisition time, by combining a train of Magnetization Transfer (MT) off-resonance saturation pulses with a small field-of-view single-shot Echo Planar Imaging (EPI) readout (ZOOM-EPI [7]). The feasibility of the approach is demonstrated in 3 healthy volunteers.

MRI Acquisition:

3 healthy volunteers (25-28 years) were imaged using a 3T Philips Achieva scanner with a 32-channel head coil and radio-frequency dual transmit technology.

MT-weighted data were acquired at 18 combinations of frequency offset and MT pulse flip angles, with 6 non-MT-weighted (M₀) images interleaved, giving an acquisition time of 24mins (details in figure 1). Imaging volume consisted of twelve 5mm-thick axial slices centred at level C2-C3, FOV=51x41mm², 0.8x0.8mm² in-plane resolution, TE=27ms, TR=10040ms (4 slices per TR), NSA=2.

A train of off-resonance pulses was applied prior to slice excitation. To avoid contamination between partial saturation in the imaging volume due to the ZOOM-EPI tilted refocus and spatially non-selective MT pulses, a delay of 6.5s was appended after each slice package. Artefact-free images were obtained without outer volume suppression, allowing slice excitation starting immediately after the MT pulse train, thereby almost entirely preserving the MT-weighting.

For T₁ estimation, Inversion Recovery (IR) data (6mins) were acquired at 8 different Inversion Times (TI=100, 220, 340, 460, 1300, 1420, 1540, 1660ms) with the same readout as MT-weighted data.

Data Analysis:

MT-weighted and IR data were co-registered to the mean of the interleaved M₀ images using slice-wise rigid transformations estimated with flirt (http://www.fmrib.ox.ac.uk/fsl/). M₀ images were also exploited to characterise the noise distribution, and to normalize the MT-weighted signal prior to qMT model fitting.

To account for the slice-dependent MT-weighting and the unmet steady-state condition introduced by the sequence, data were fitted using the Minimal Approximation Magnetization Transfer (MAMT) model [7], implemented in Matlab (The MathWorks Inc., Natick, 2000) using a discretizing step of 120µs. Four model parameters were estimated: the bound pool fraction (BPF), free water pool transverse relaxation time (T₂^F), bound water pool transverse relaxation time (T₂^B) and the forward MT exchange rate (RM₀^B). A super-Lorentzian lineshape was assumed to describe the bound pool saturation rate. The free water pool longitudinal relaxation rate (R₁^F) was obtained via mono-exponential model fitting of IR data, assuming a bound water pool longitudinal relaxation rate (R₁^B) of 1s^-1 [9]. Maximum likelihood estimation based on Rician noise was used.

Figure 2 shows single slice examples of IR and MT-weighted images used to estimate model parameters. Figure 3 gives single voxel examples of model fitting for the IR and MT experiments. Single slice BPF, T₂^F, T₂^B, RM₀^B and T₁ maps are shown in figure 4 together with the averaged M₀ image.

Whole cord parameters median values and interquartile ranges are reported in figure 5. Global mean and standard deviations (SD) were: BPF=10.5(±0.18)%, T₂^F=47.5(±3.8)ms, T₂^B=10.1(±0.27)µs, RM₀^B=1.73(±0.09)s^-1 and T₁^obs=1130(±36.3)ms.

Median values reported in figure 5 are consistent with previous findings in brain studies [10], suggesting that estimation of two-pool qMT model parameters is feasible with this approach. Visual inspection of parametric maps and parameter values in figure 4 shows that whole cord distributions for T₂^F and RM₀^B are broader than for BPF and T₂^B, confirming previous findings using the same model in the brain [8,10], and outcomes of protocol optimisation [11].

In future work a separate acquisition could be used to better estimate T₂^F , as in [3], and B₁ and B₀ corrections should also be implemented.

This novel approach for in vivo qMT in the SC using rapid single-shot ZOOM-EPI readout immediately following a train of MT pulses gives improved protocol flexibility, since the MT saturation and image acquisition can be separately designed. The requirement for high resolution data therefore does not interfere with the amount of MT-weighting, which can be optimally designed to achieve time efficiency.