Highly Accelerated EPI Evaluated for fMRI

Introduction Multiplexed EPI, a type of simultaneous multi-slice imaging, utilizes multiband (MB) rf pulses and simultaneous image refocusing (SIR) to reduce the sampling TR of resting state and task fMRI [1-2]. Using s SIR refocusings and m images per MB excitation, the total number of images in each echo train is N=s×m. Here we evaluate different values of s and m with up to N=48 images per EPI echo train. Evaluations were quantified with temporal SNR (tSNR), BOLD CNR and movie-frame classification accuracy in fMRI studies. Methods Subjects were scanned on a Siemens 3T Trio with a 32 channel coil. Parameters: resolution= 2.5 x 2.5 x 3 mm, TE=36 ms (for SIR s=2 the TEs are 36ms and 41ms; for SIR s=3, TEs 36ms, 41ms, 46 ms), whole brain coverage and controlled aliasing = FOV/4 shift [3]. RF duration was 5.2ms (for higher m, it is lengthened to up to 10ms due peak power limitations). The effects of s and m (s=1-3 and m=1-16) on tSNR (averaged across all voxels in slices acquired with TE=36ms) were evaluated on resting subjects while holding all other scan parameters constant at TR=500ms. Based on the tSNR results, accelerations were limited to N<=16 in subsequent BOLD evaluations. In visual stimulation studies, 5 subjects fixated while watching a 15s on 15s off 4Hz flickering checker board pattern 9 times per combination of s and m. In classification fMRI studies, recent developments enabled TE to be constant across SIR slices when s=2 [4]. Subjects fixated while watching a 36 second movie repeated 9 times per combination of s and m at near minimum TR. The information content of the resulting BOLD signals was measured using leave-one-repeat-out classification [5]. The number of possible frames for each N was 36 s / TR used. Results Fig 1 shows increases in s and m factors generally decreased tSNR. However for constant N, s=2 generally has higher tSNR than s=1 especially for high total accelerations (N > 8). Fig 2 shows better image quality with s2m8 than with s1m16. This is expected given g-values increase with m but not s. With s2, dropout is slightly worse due to longer echo train. Fig 3 shows the comparison of the mean t-value and number of voxels with t>2.5 (p<0.01, uncorrected). Fig 4 shows the number of timepoints classified for acceleration factors 1 to 16 in the movie experiment. N=8-16 with s=2 (TR=300-600ms) gave the largest number of time points classified.