As compared to conventional whole-body MRI, a dedicated head-only MRI gradient system^1,2 can provide simultaneously high gradient amplitude (G_max=80 mT/m) and high slew rate (SR=700 T/m/s) without limitations of whole-body peripheral nerve stimulation^3-4. While high Gmax is beneficial for reducing echo times (TE) in diffusion imaging, the high SR allows EPI readout echo-spacing (ESP) to be halved, which provides the dual benefit of reduced EPI distortion and further reduction of TE due to shortened echo-train-length. These benefits are critical to diffusion microstructure imaging, which as compared to conventional DWI/DTI pulse sequences, require higher b-values and longer diffusion intervals (Δ) that lengthen TE and reduce SNR. It is therefore of interest to explore the benefits to diffusion microstructure imaging from this high-performance MRI system.

Two normal subjects (under an IRB-approved protocol) underwent diffusion imaging using the dedicated head-only gradient coil at 3T, and again using conventional 3T MRI (GE MR750), using the same 32-channel brain coil (Nova Medical, Wilmington MA). Single spin-echo diffusion preparation with the minimum possible TE was utilized for axial brain imaging at b-values of 1000, 3000, and 8000 s/mm².

An AxCaliber⁵ sampling scheme was also acquired using only the head-only system with 4 diffusion intervals of Δ={24, 32, 39, 51} ms and 8 evenly-spaced diffusion-encoding gradient amplitudes of G=10-80 mT/m in the superior-inferior direction (FOV=21 cm, 2 mm-isotropic sampling, partial-ky=0.75, 2 signal averages, TR/TE=3000/85 ms, gradient pulse width δ=9.5 ms). No parallel imaging and EPI-distortion correction were applied. To provide orientation information, DTI sampling scheme was also acquired with b-values of {1,000, 2,000, 12,000} s/mm² keeping constant Δ=51 ms, TE=85 ms (30 gradient directions).

Multi-compartment-model fitting was performed in the corpus callosum assuming two compartments (cylinder-zeppelin⁶) to obtain axonal radius, axonal fraction (restricted), and orthogonal diffusivity. The parallel diffusivity of the zeppelin compartment (hindered/extra-axonal) and the diffusivity of the cylinder compartment (restricted/intra-axonal) were fixed to the parallel diffusivity as obtained by a DTI fit. A least-squares fitting (formulation) was used, initialized using a grid-search (grid size of 10) and optimization via the interior-point algorithm. Fitting was performed with both the AxCaliber and DTI data. To evaluate the effect of data at high b-value, fitting was performed by incrementally adding DTI data at b-values of {1,000, 2,000, 12,000} s/mm².

The head-only system allowed the use of higher EPI readout gradient amplitude of 45 mT/m compared to 24 mT/m of the conventional whole-body system, effectively halving ESP (Table 1). As SNR would be reduced due to increased EPI readout amplitude, a slightly increased EPI readout gradient amplitude was used (29 mT/m) resulting in a reduction in ESP by a third. Both head-only images had visibly reduced susceptibility effects (Figure 1). This reduction in ESP that reduced echo-train-length accounted for approximately 6-9 ms reduction in TE regardless of diffusion b-value. The net TE reduction was between 18 ms and 28 ms, which resulted in visibly higher white-matter SNR between 9% and 29% (assuming T2=80 ms). As a result of reduced TE, suppression of subcutaneous fat and tissue was also reduced.

Imaging at the longer TE=85 ms provided adequate SNR (Figure 2), even at b=12,000 s/mm² where good white matter contrast was observed despite Rician noise. The microstructure metrics in the corpus callosum were averaged at five points along the corpus callosum in the anterior (genu) to posterior (splenium) direction. The axonal radius was on average between 5-6 µm and was not significantly different between the two subjects (Figure 3). The addition of the b=12,000 s/mm² reduced the variance of both axonal radius and orthogonal diffusivity (Figure 4). There was a disparity of about 0.2 in the axonal fraction of both subjects, which indicates that the parameters could be better optimized and the model better conditioned. There was high residual fat aliasing, which could improve with better fat suppresion⁷.

In our preliminary investigation of in-vivo microstructure imaging using the head-only gradient coil in-vivo scanner, the diffusion scans provided adequate image quality at 2 mm-isotropic spatial resolution. The scans were also well-tolerated with no peripheral nerve stimulation reported. The axonal radii were over-estimated, a result that was similar to that observed in other work⁸ and that would improve with the availability of better models and an even higher gradient amplitude for diffusion-encoding. The reducd TE and ESP provided higher white matter SNR and reduced EPI distortion without the use of parallel imaging and EPI-distortion correction techniques.