Converting Multi-Shell and Diffusion Spectrum Imaging to High Angular Resolution Diffusion Imaging

Fang-Cheng Yeh,Timothy D Verstynen

doi:10.3389/fnins.2016.00418

Abstract

Multi-shell and diffusion spectrum imaging (DSI) are becoming increasingly popular methods of acquiring diffusion MRI data in a research context. However, single-shell acquisitions, such as diffusion tensor imaging (DTI) and high angular resolution diffusion imaging (HARDI), still remain the most common acquisition schemes in practice. Here we tested whether multi-shell and DSI data have conversion flexibility to be interpolated into corresponding HARDI data. We acquired multi-shell and DSI data on both a phantom and in vivo human tissue and converted them to HARDI. The correlation and difference between their diffusion signals, anisotropy values, diffusivity measurements, fiber orientations, connectivity matrices, and network measures were examined. Our analysis result showed that the diffusion signals, anisotropy, diffusivity, and connectivity matrix of the HARDI converted from multi-shell and DSI were highly correlated with those of the HARDI acquired on the MR scanner, with correlation coefficients around 0.8~0.9. The average angular error between converted and original HARDI was 20.7° at voxels with signal-to-noise ratios greater than 5. The network topology measures had less than 2% difference, whereas the average nodal measures had a percentage difference around 4~7%. In general, multi-shell and DSI acquisitions can be converted to their corresponding single-shell HARDI with high fidelity. This supports multi-shell and DSI acquisitions over HARDI acquisition as the scheme of choice for diffusion acquisitions.

Highlights

Diffusion MRI offers a non-invasive way to map the structural connectivity of the human brain (Behrens et al, 2003, 2007; Wedeen et al, 2012), and several diffusion sampling schemes have been used to acquire the diffusion MRI data
The high correlation coefficient (>0.9) suggests that the converted high angular resolution diffusion imaging (HARDI) is a good predictor of the original HARDI
We found that the fiber orientation distribution function (fODF) from the converted HARDI data are consistent with the fODFs from the original HARDI data

Summary

Introduction

Diffusion MRI offers a non-invasive way to map the structural connectivity of the human brain (Behrens et al, 2003, 2007; Wedeen et al, 2012), and several diffusion sampling schemes have been used to acquire the diffusion MRI data. MRI data and facilitate comparing and aggregating results across studies with different acquisition approaches. To explore this possibility, we used generalized q-sampling reconstruction (Yeh et al, 2010) to interpolate multi-shell and DSI data into their corresponding HARDI representation. We used generalized q-sampling reconstruction (Yeh et al, 2010) to interpolate multi-shell and DSI data into their corresponding HARDI representation These converted data sets are compared with the original HARDI data acquired using a single-shell scheme. This comparison was conducted in both a single phantom study and several in vivo human studies. The network measures (Bullmore and Sporns, 2009) were calculated using graph theoretical analysis to examine their difference

Methods

Results

Conclusion