“Windowed” Composite Reconstruction Improves Rotating Short-Axis High-Resolution DWI (RSA-DWI) in both Simulation and Human data
Qiuting Wen1, Chandana Kodiweera2, and Yu-Chien Wu1

1Radiology and Imaging Sciences, Indiana University, Indianapolis, IN, United States, 2Darmonth College, Hanover, NH, United States

Synopsis

High-resolution DWI often relies on multi-shot acquisitions, which suffer from long acquisition time and motion-related phase issues. However, highly correlated information exists in DWIs as they are weighting the same structure. To take advantages of this feature, rotating short-axis DWI was proposed to accelerate DWI acquisition by acquiring only one rotating blade per diffusion direction. In the previous reconstruction, high-resolution DWI was achieved by integrating the full set of DWIs. In this work, we propose a “windowed” composite reconstruction where only a subset of DWIs was selected to reconstruct each high-resolution DWI. Improved image quality was appreciated in both simulation and human data.

Introduction

High-resolution whole brain diffusion tensor imaging (DTI) with an in-plane resolution of 1×1mm2 or higher is impractical with single shot echo planar imaging (SS-EPI) due to the prolonged TE and severe geometric distortion, as the k-space is four times bigger than the conventional 2×2mm2. Alternative approaches are mainly multi-shot techniques [1-3], where multiple excitations are needed to reconstruct one full k-space. The prolonged acquisition time and motion related phase shifts are major issues that make their clinical application difficult. Hence, a rotating short-axis diffusion weighted imaging (RSA-DWI) acquisition method was proposed where only one blade is acquired per diffusion direction to accelerate the acquisition [4]. A highly constrained backprojection method [5] was employed to reconstruct the high-resolution image using all the DWIs (i.e., full composite reconstruction) to take advantages of the highly correlated DWIs. In this work, we improved upon the full composite reconstruction in RSA-DWI by using a subset of DWIs that are most correlated called “windowed” composite reconstruction. We showed that image quality is greatly enhanced in both simulation and in human data.

Methods

RSA-DWI: One SS-EPI blade with short axis in the frequency encoding direction was acquired per DW direction. The EPI blade rotated as DW direction changed (Figure 1B: upper and middle row). An arbitrary DW encoding scheme could be used in RSA-DWI. However, in order to optimize the correlation of each set of DWIs used to reconstruct each high-resolution DWI, diffusion directions were sorted into a sequence that was closest to an Archimedean spherical spiral curve and the blades were rotated in this order (Figure 1A). Thus, for each high-resolution DWI to be reconstructed, the selected DWIs are not far away on the sphere to guarantee similarity.

Windowed Composite Reconstruction: Compared to full composite reconstruction, only a subset of DWIs whose directions were closest to the current DWI on the sphere were used (Figure 1A highlighted in red). The k-space blades of the selected DWIs were Fourier transformed into the image space to be aligned and was then combined in the k-space to form a composite image. The high-resolution DWI was then obtained by training the low-resolution DWI using the composite image, as described in [5] and [6].

Simulation with Tensor Phantom: A 1mm isotropic brain DTI were generated from high quality brain DWIs acquired with multi-shot read-out segmented EPI sequence (RESOLVE) at a 3.0T Siemens Prisma scanner at b=1000s/mm2. The tensor matrix was used as a phantom for DWI simulation and was chopped in the k-space to form RSA-DWI (Figure 1B: middle row). 61 RSA-DWIs were generated with blade size of 32×256 and rotation angle of 15° using the direction ordering scheme described above. Both full and windowed composite method (window size = 12) were applied to reconstruct the data for comparison. Figure 1&2A demonstrated this scheme.

Human data: RSA-DWI was performed on a healthy volunteer at a 3.0T Philips Achieva INTERA scanner with an 8-channel head coil with 1mm isotropic resolution, ten slices, 61 diffusion directions and b-value = 1000s/mm2 at TE = 122ms, TR = 1800ms, scan time = 4’45’’. Blade size and rotation angle were the same as in the simulation. b0s were acquired twice with opposite phase encoding directions and were used to correct for geometric distortions prior to image reconstruction (TOPUP, FSL).

Results

In the simulation with only 1/8 of the full k-space data, windowed composite reconstruction demonstrated its ability of recovering high-resolution DW images by exploring the correlations of its neighboring DWIs on the sphere (Figure 1C). FA maps showed great improvement compared to full composite reconstruction especially in fine fiber regions (Figure 2A). Such improvement could also be appreciated in the human data (Figure 2B).

Discussions

RSA-DWI is a fast high-resolution diffusion imaging technique where only one EPI blade is acquired per diffusion direction, accelerating the acquisition by ~10 fold compared to the state-of-art multi-shot EPI techniques. High-resolution DWIs are reconstructed through windowed composite method by exploring correlations in neighboring DWIs. Although DTI results were presented in this work, RSA-DWI can be applied to any diffusion analyses including parametric diffusion modeling and non-parametric q-space approaches. Furthermore, the more diffusion directions the better the windowed composite reconstruction works, as when the windowed DWI directions are closer on the sphere they become more correlated. This short-axis EPI based acquisition inherits its advantage of reduced distortion and ghosting compared to long-axis EPI [2]. And the distortion was further corrected in our application by acquiring a set of b0s with reversed phase-encoding gradients. Compared to simulation, data quality was compromised in human data mainly due to noise contamination, eddy current and prolonged TE (122ms vs. 67ms in simulation). Methods to reduce TE in conventional EPI can be applied in RSA-DWI such as partial k-space sampling and parallel imaging (e.g. SENSE or GRAPPA). These could further improve SNR and will be implemented in future studies.

Acknowledgements

This research is supported by Dartmouth Synergy pilot grant and Indiana Alzheimer Disease Center (iADC) pilot grant.

References

[1] Porter DA, Heidemann RM. High resolution diffusion-weighted imaging using readout-segmented echo-planar imaging, parallel imaging and a two-dimensional navigator-based reacquisition. Magn Reson Med. 2009 Aug;62(2):468-75.

[2] Skare S, Newbould RD, Clayton DB, Bammer R. Propeller EPI in the other direction. Magn Reson Med. 2006 Jun;55(6):1298-307.

[3] Forbes KP, Pipe JG, Karis JP, Heiserman JE. Improved image quality and detection of acute cerebral infarction with PROPELLER diffusion-weighted MR imaging. Radiology. 2002 Nov;225(2):551-5.

[4] Wu YC , Mistretta C. A. Alexander A. L. et al. High Angular Resolution Diffusion Imaging (HARDI) with Highly Constrained Back Projection Reconstruction (HYPR). ISMRM 2010. P1614.

[5] Wu YC, Kodiweera C. Rotating short-axis EPI “Blades” as veering diffusion gradient directions with composite reconstruction (RSA). ISMRM 2014. P0666.

[6] Mistretta CA, Wieben O, Velikina J, Block W, Perry J, Wu Y, Johnson K, Wu Y. Highly constrained backprojection for time-resolved MRI. Magn Reson Med. 2006 Jan;55(1):30-40.

Figures

A. Sorting scheme of diffusion directions on a sphere. Blades were rotating as DW direction changed in this order. Windowed DWIs (red) were selected to reconstruct DWI-1 (green). B. Rotating short-axis blades (middle row) and corresponding low-resolution DWIs used to reconstruct DWI-1 (simulated data). C. Ground truth (above) and reconstructed DWI-1 using “windowed” composite method (below).

A. FAs generated using “full” composite reconstruction (left) and “windowed” composite reconstruction (middle) with simulated data. Difference images were generated by subtracting the ground truth FA map (right). B. FAs reconstructed with human data using the two reconstruction methods. Color coded major eigenvectors were shown below.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
0516