Functional Magnetic Resonance Imaging at High Spatiotemporal Resolution using EPI Combined with Different k-Space Undersampling Techniques at 3 Tesla

Abstract

In functional magnetic resonance imaging (fMRI), major drawbacks of the commonly used echo-planar imaging (EPI) sequence are limited spatial specificity due to blurring and distortions as well as signal cancellation in areas affected by susceptibility gradients, such as the orbitofrontal cortex (OFC). In contrast, segmented EPI techniques facilitate ultra-high spatial but low temporal resolution. In this work, an EPI sequence with optimized slice-dependent echo time was developed avoiding signal drop outs in the OFC in 50 % of all subjects during fMRI (N = 12) compared to a standard EPI sequence. The average number of activated voxels detected in the OFC was thereby increased by a factor of 6.3. It was further shown for the first time that the spatial specificity in EPI fMRI at 3 T can be improved by increasing the matrix size in combination with the parallel imaging factor beyond conventional EPI parameter settings. By using the proposed high-resolution compared to a standard EPI protocol, the multi-subject analysis of a simple fingertapping task (N = 6) and a sophisticated motivation task (N = 15) showed robust and clearly less blurred activation in the sensorimotor cortex (SMC) and in the nucleus accumbens (NAcc), respectively. The number of separable clusters detected in the SMC and in the NAcc was thereby increased by a factor of 2.7 and 1.4, respectively. In order to perform fMRI at ultra-high spatial and high temporal resolution, a segmented EPI sequence was highly accelerated (R = 8) with the so-called UNFOLD technique. Both, the MR sequence and data post-processing were optimized facilitating the robust detection of neuronal activation at 0.7 x 0.7 mm2 resolution and half-brain coverage. Last but not least, a novel filtering strategy is proposed minimizing temporal coherences in UNFOLD datasets and thus improving the detectability of neuronal activation. By using the proposed filter compared to a standard filter, the number of activated voxels detected in the SMC (N = 5) was increased up to a factor of 1.4.