Q-Space Truncation and Sampling in Diffusion Spectrum Imaging

Purpose: To characterize the q-space truncation and sampling on the spin-displacement probability density function (PDF) in diffusion spectrum imaging (DSI). Methods: DSI data were acquired using the MGH-USC connectome scanner (G(max) = 300 mT/m) with b(max) = 30,000 s/mm(2), 17 x 17 x 17, 15 x 15 x 15 and 11 x 11 x 11 grids in ex vivo human brains and b(max) = 10,000 s/mm(2), 11 x 11 x 11 grid in vivo. An additional in vivo scan using b(max) = 7,000 s/mm(2), 11 x 11 x 11 grid was performed with a derated gradient strength of 40 mT/m. PDFs and orientation distribution functions (ODFs) were reconstructed with different q-space filtering and PDF integration lengths, and from down-sampled data by factors of two and three. Results: Both ex vivo and in vivo data showed Gibbs ringing in PDFs, which becomes the main source of artifact in the subsequently reconstructed ODFs. For down-sampled data, PDFs interfere with the first replicas or their ringing, leading to obscured orientations in ODFs. Conclusion: The minimum required q-space sampling density corresponds to a field-of-view approximately equal to twice the mean displacement distance (MDD) of the tissue. The 11 x 11 x 11 grid is suitable for both ex vivo and in vivo DSI experiments. To minimize the effects of Gibbs ringing, ODFs should be reconstructed from unfiltered q-space data with the integration length over the PDF constrained to around the MDD. (C) 2016 International Society for Magnetic Resonance in Medicine