Conventional multi-shot DWI is known to suffer from inter-shot phase inconsistencies due to motion and hardware imperfections. In this work, we present a new approach for phase-error-free diffusion imaging by using multi-shot Paddlewheel-shaped EPI acquisition and filtered back projection (FBP) Reconstruction. The necessity of inter-shot phase correction is removed due to the magnitude-only nature of FBP. Reduced FOV excitation scheme is incorporated to reduce scan time and artifacts. Exemplary results of head and prostate DWI are demonstrated to show the efficacy of the proposed methods.
Acquisition: a 3D Planes-on-a-Paddlewheel (POP) EPI readout sampling (Fig 1) was developed for stack-of-stars radial acquisition. During each shot, the phase encoding blips were placed on the slice direction and the sampling trajectory traversed in a standard Cartesian EPI manner. The data acquisition plane then rotated along kz axis until the whole 3D k-space was filled. The double refocus method with bipolar diffusion gradients10 was used to minimize the eddy currents induced by the diffusion gradients. EPI reference scans were also acquired at each azimuthal angle to correct errors caused by gradient delays, B0 inhomogeneity and eddy currents.
Reduced FOV: A reduced FOV excitation scheme was also incorporated to improve radial sampling efficiency. The slice selection gradient lobes were placed at three different axes to excite a volume. Crusher gradients were applied to spoil any unwanted signals.
Gradient Delay: Gradient delays were measured on a sphere phantom via Peters’ method11.The radial trajectories were corrected for gradient delays by addition a constant timing compensation for Gx & Gy.
Reconstruction: Fig 2 shows the reconstruction flow. First, the reference scan data were used to perform intra-shot phase correction as a standard EPI pre-processing procedure. The data from each shot was then Fourier transformed into the projection space perpendicular to the acquisition plane. Finally, 2D FBP was used to transform the projection data back to image domain for each slice. The whole reconstruction process was performed on a PC with Inter i7@3.6GHz and 32GB RAM using Matlab.
The proposed method was implemented on a 1.5T whole-body scanner (Centauri, Alltech Medical Systems), with gradient strength 33mT/m and slew rate 130mT/m/s. Head and prostate DWI data were acquired. Head DWI parameters: FOV 200*200*90mm3, 10% slice-oversample, 1.3mm isotropic resolution, 3 diffusion directions, b value 0/800, 240 spokes per stack, FA/TR/TE 60/500ms/77.6ms, NEX 1 and total scan time 600s. Prostate DWI parameters: FOV 100*100*60mm3, 10% slice-oversample, 1mm isotropic resolution, 3 diffusion directions, b value 0/800, 150 spokes per stack, FA/TR/TE 60/500ms/75.6ms, NEX 1 for b0, NEX 2 for b800 and total scan time 600s.
Head DWI results from a healthy volunteer is demonstrated in Fig 3. No visible signal corruption, artifact or noise amplification as a result of inter-shot phase variations can be seen. The Image intensity inhomogeneity is due to the uncorrected phase array coil sensitivity variation.
Fig 4. Shows representative results of the prostate DWI and ADC map from a healthy volunteer. Reducing the FOV lead to decreased scan time and enabled high-resolution acquisition of the prostate without visible susceptibility or motion artifacts.
The proposed approach can provide robust, high SNR 3D diffusion imaging without any inter-shot phase inconsistency issues, which mainly benefits from the radial sampling scheme and FBP reconstruction. Compared to the conventional multi-shot DWI techniques that require complex inter-shot phase correction mechanisms, the computational cost is trivial. In our case, the reconstruction time for a matrix size of 256*256*80 is less than 10 second via Matlab with single thread.
Slight blurring can be observed in the results, which we suspect is due to the inaccurate radial trajectories. This is an inherent problem in radial imaging, and in particular with the EPI readout scheme. Efforts will be made to further improve the sampling trajectory accuracies in the future work.
Radial sampling is inherently less efficient than Cartesian sampling. With the reduced FOV excitation, the sampling efficiency is improved but the total scan time is still long. Further scan acceleration can be achieved by incorporating model-based reconstruction techniques like compressed sensing.