High‐fidelity, high‐spatial‐resolution diffusion magnetic resonance imaging of ex vivo whole human brain at ultra‐high gradient strength with structured low‐rank echo‐planar imaging ghost correction

Abstract

Diffusion magnetic resonance imaging (dMRI) of whole ex vivo human brain specimens enables three‐dimensional (3D) mapping of structural connectivity at the mesoscopic scale, providing detailed evaluation of fiber architecture and tissue microstructure at a spatial resolution that is difficult to access in vivo. To account for the short T2 and low diffusivity of fixed tissue, ex vivo dMRI is often acquired using strong diffusion‐sensitizing gradients and multishot/segmented 3D echo‐planar imaging (EPI) sequences to achieve high spatial resolution. However, the combination of strong diffusion‐sensitizing gradients and multishot/segmented EPI readout can result in pronounced ghosting artifacts incurred by nonlinear spatiotemporal variations in the magnetic field produced by eddy currents. Such ghosting artifacts cannot be corrected with conventional correction solutions and pose a significant roadblock to leveraging human MRI scanners with ultrahigh gradients for ex vivo whole‐brain dMRI. Here, we show that ghosting‐correction approaches that correct for either polarity‐related ghosting or shot‐to‐shot variations in a separate manner are suboptimal for 3D multishot diffusion‐weighted EPI experiments in fixed human brain specimens using strong diffusion‐sensitizing gradients on the 3‐T Connectom MRI scanner, resulting in orientationally biased dMRI estimates. We apply a recently developed advanced k‐space reconstruction method based on structured low‐rank matrix (SLM) modeling that handles both polarity‐related ghosting and shot‐to‐shot variation simultaneously, to mitigate artifacts in high‐angular resolution multishot dMRI data acquired in several fixed human brain specimens at 0.7–0.8‐mm isotropic spatial resolution using b‐values up to 10,000 s/mm2 and gradient strengths up to 280 mT/m. We demonstrate the improved mapping of diffusion tensor imaging and fiber orientation distribution functions in key neuroanatomical areas distributed across the whole brain using SLM‐based EPI ghost correction compared with alternative techniques.

Eddy current-induced artifacts correction in high gradient strength diffusion MRI with dynamic field monitoring: demonstration in ex vivo human brain imaging

“…The solution can be obtained in closed form with the Moore-Penrose inverse matrix, which models the linear relationship between x and measurements y k z ,s,p . 35 However, this strategy was not feasible to achieve in practice because of the large dimensionality of the data at hand. The following assumption made the problem tractable.…”

“…We have used this assumption before with satisfactory results. 35 We then applied a non-uniform inverse Fourier Transform along the partition direction k z , 62 and treated the problem as a 2D multi-shot reconstruction approach for each slice index z . We adjusted the image encoding matrix E k z ,s by: 1) removing the contribution of c 3 ( t ) to the phase model, and 2) modifying the grid r m to account for a fixed slice z .…”

“…2D MUSE was applied right after a non-uniform inverse Fourier transform along the slab encoding. Our previously published method that extends the structured low-rank matrix-based method 36 to a multi-shot acquisitions and handles both nonlinear spatial variations in phase between odd and even echoes and between shots simultaneously (SLM-based method). 35 As with 2D MUSE, the k-space data was first non-uniform inverse Fourier transformed along the slab encoding. Shots and odd/even images were combined with rSoS.…”

“…35 The residual ghosting artifacts bias subsequent dMRI analyses in a diffusion-direction dependent manner. 35 Advanced k-space reconstruction methods that simultaneously handle nonlinear spatial phase variations between odd and even echoes and shots within an undersampled image reconstruction framework can effectively mitigate ghosting artifacts in high-gradient strength dMRI. 35,36 Nevertheless, these methods do not explicitly incorporate information on the spatiotemporal phase evolution due to eddy currents in the image reconstruction model.…”

“…35 Advanced k-space reconstruction methods that simultaneously handle nonlinear spatial phase variations between odd and even echoes and shots within an undersampled image reconstruction framework can effectively mitigate ghosting artifacts in high-gradient strength dMRI. 35,36 Nevertheless, these methods do not explicitly incorporate information on the spatiotemporal phase evolution due to eddy currents in the image reconstruction model. They may underperform under certain circumstances, e.g., when the undersampled reconstruction problem becomes harder to solve due to increasing number of shots but limited coil sensitivity encoding in highly segmented, high-spatial resolution EPI acquisitions.…”