Multiband Multiecho (MBME) EPI has the advantage of higher temporal resolution and improved sensitivity to BOLD signal compared to conventional single band single echo EPI. A few studies have examined the benefits of MBME in resting state and task fMRI [1,2]. It has also been demonstrated that TE-dependent analysis methods (e.g Multiecho Independent Component Analysis ME-ICA [3]) can be applied to MBME data to effectively remove non-BOLD signals [1]. However, we recently observed differences in the slow signal drift between multiple echoes in MBME data, which could compromise the TE-dependence model and impact the ME-ICA analysis. Figure 1C. shows an example of such drift differences in an MBME data acquired on an agar phantom. For better illustration, we show data from an acquisition with a significant amount of aliasing artifact in the reconstructed images (Figure 1A). We have noted that areas of aliasing artifacts often exhibit large drift differences. Our goal is to design a metric to quantify the differences in drift among multiple echoes, and to probe the effects of the drift differences on TE dependent analysis such as ME-ICA. The previous MBME studies have used in-plane acceleration for faster data acquisition and reduced off resonance effects. Gomez et al [4] also indicated that in-plane acceleration combined with lower multiband factor (2 or 3) increases BOLD temporal SNR. In this study, we acquired both phantom and in-vivo MBME data with and without in-plane acceleration (Auto-calibrating Reconstruction for Cartesian Imaging: ARC) so we can compare the effects of in-plane acceleration on the signal drift. We used a GE 3T Discovery MR750 (GE Healthcare, Waukesha, WI) and a prototype MBME pulse sequence with blipped CAIPI [5]. The parameters were: MB=3, 250 time points, 3 echoes, 24cm FOV, and 3.75x3.75x3.8mm voxel size. For data without ARC: TR = 1.2s, TE = 11, 32.2, and 53.4ms; and with ARC=2: TR = 1s, TE = 11, 25, and 39ms. An offline Matlab recon was used for image reconstruction. To quantify the drift differences, we first performed a fit of the MBME data to obtain T2* and S0 time courses (Figure 2). The residual time course from this fit was then modeled voxel-wise as a 10th order Legendre polynomial to remove high frequency noise components that did not contribute to slow drifts. From this fitted residual, we calculated the standard deviation and used it as a quantitative measure of the drift difference (dubbed as MEQA, Figure 2.) For the in-vivo data, we also performed ME-ICA analysis.

Figure 1B shows an example of the MEQA map. High MEQA values correspond well to voxels with remaining aliasing artifacts and large echo drift differences. Figure 3 shows calculated MEQA maps from the MBME phantom and in-vivo data. The data with in-plane acceleration (ARC=2) has significantly more MEQA voxels than the data without ARC, indicating more prevalent deviation from the TE dependence model in the ARC=2 data. This is true for both the phantom and in-vivo data.

The ME-ICA analysis shows that the data acquired without ARC has higher number of both total ICA components and BOLD-like components than the data with ARC (Table 1.)

Our results show that the MEQA metric is a sensitive measure of slow drift differences among multiple echoes in MBME data. Such differences are typically observed in voxels with imaging artifacts due to imperfect image reconstruction, susceptibility, and motion (not shown). Therefore MEQA could potentially be used as a quality assurance metric for MBME data. We also show that data with higher MEQA voxel counts, and thus more prevalent echo drift differences, has lower ME-ICA performance. Although very preliminary, this result warrants further investigations of the source of the drift differences and how they affect ME-ICA analysis.