Accelerated 3D-EPI fMRI Using Parallel Imaging

Introduction: Fast, three-dimensional acquisition is advantageous for fMRI1. While advanced 3D methods have been demonstrated for fMRI, 2Dmulti- slice EPI remains the convention. Extension to 3D-EPI using phaseencoding in the slice selection direction has presented feasible results2, but to our knowledge so far only within the conventional EPI temporal framework. Parallel imaging (PI) can provide a desirable improvement in both temporal resolution and CNR. This work seeks to evaluate some features of employing accelerated 3D-EPI for fMRI compared to conventional method. Methods: Data acquisition: 3T Siemens scanner, 8-channel head coil, segmented EPI sequence. Studies were performed on a healthy volunteer using four EPI schemes: conventional 2D-multi-slice, 3D full volume scan, 3D scans accelerated in 1-dimension with reduction factors R=2 and R=4. All acquisition parameters were identical except for variations inherent to the 2D sequence. Volumes of 20 slices were attained in ~7 s in 2D-multislice and down to ~2 s in 3D with R=4. Reconstruction was offline using GRAPPA. Functional imaging: motor task paradigm with self-paced, right-handed finger tapping ~30s off / 30s on blocks. Time series with 90 repetitions were acquired and activation maps (t-scores) were generated with variations in thresholds accounting for differences in intrinsic CNR and SNR (fig. 2). Results: Results show overall similar activation structure in the contralateral primary motor cortex and the supplementary motor area (fig. 1). Activation was detected consistently with all acquisition schemes (fig. 2), even with high 1-dimensional undersampling. The activation maps for R=4 showed a smaller reduction in t-scores (~30 ) compared to the reduction in SNR (75). Further, acceleration reduced total scan time up to a factor of 3.5 relative to the full k-space acquisitions. Conclusion: Acceleration offers significant gains to 3D-EPI for fMRI. Although loss of spatial SNR with shortened acquisition time is expected to reduce t-scores, acceleration is feasible due to the increase in acquired volumes per time and relatively disproportional smaller loss of CNR. Limits to acceleration are indicated in the activation maps as further reduction will give proportionally decreased CNR. However, high 1-dimensional reduction factors were shown to be feasible, achieving both spatial specificity of the functional response and higher temporal resolution than obtainable with 2Dmulti- slice within equal scan durations and spatial coverage. Additionally, a 3D scheme offers the possibility of two-dimensional acceleration for further imaging flexibility