Simultaneous multi-slice echo-planar imaging (EPI)¹, also known as multiband (MB) EPI², has recently gained much attention for its ability to enhance spatiotemporal resolution in resting-state functional MRI (rs-fMRI). However, effects arising from respiration and cardiac pulsatility as well as respiration volume per time (RVT)³ and cardiac rate variation (CRV)⁴, commonly referred to as "physiological noise” in rs-fMRI⁵, affect MB-EPI signal quality in a manner that has yet to be well characterized. In particular, residual aliasing has the potential to occur in MB slice acceleration and to introduce spurious signals from one slice into another in rs-fMRI data. Because a given group of slices samples the physiological noise at the same time, it is hypothesized that physiological noise effects are amplified in MB-EPI versus regular EPI. Experiments are conducted to verify this hypothesis and also to investigate strategies that can correct for this problem.

Twelve participants were each scanned at rest with both an MB and a regular EPI sequence with matching parameters. Single-shot MB-EPI was acquired by gradient-echo (GE)-EPI² (TR=323ms, TE=30ms, flip angle=40°, 15 slices, 3.44mm×3.44mm×4.6mm, acceleration factor=3, phase encoding shift factor=2). Also, a non-accelerated GE-EPI (a.k.a. “regular EPI”) was acquired with the same parameters except for the number of slices=7 (prescribed to overlap slices 9-15 of the MB-EPI). In all analysis, the first 8 slices of MB were ignored. For each participant, the respiratory and cardiac signals were recorded.

To quantify physiological effects for both rs-fMRI datasets, following retrospective motion and slice-time correction, the voxel-wise correlation was computed between the time series and the a) respiratory, b), cardiac, c) RVT and e) CRV signals. The spectral energy of the time course of each voxel was also calculated in the e) respiration and f) cardiac frequency range determined specifically for each subject, over a bandwidth of 0.15Hz. The relatively low TR value allowed the first harmonics of both respiration and cardiac processes to be captured well. Additionally, all the physiological noise metrics (a-f) were obtained after applying 1) no correction; 2) image-based retrospective correction (RETROICOR)⁶; 3) RETROICOR+RVTcor; 4) RETROICOR+CRVcor; and 5) RETROICOR+RVTcor+CRVcor, where RVTcor and CRVcor refer to inclusion of the respective physiological noise signals in the correction model. For both data sets, the spatial average of the absolute values of the six (a-f) physiological noise metrics under the five correction schemes (1-5) was calculated for each participant, from which the group mean and standard deviation were obtained.

Fig. 1 represents the group mean and standard deviation of the spatial average of the six (a-f) physiological noise metrics obtained under the five correction schemes. A mixed pattern of effects was observed. Without any correction, MB-EPI was more sensitive to physiological noise effects than regular EPI on three metrics: elevated mean correlation with the CRV signal; and elevated mean of RVT and respiratory spectral energy with increased intra-group variability. The other three metrics showed equivocal or opposite results. MB-EPI and regular EPI showed similar correlations with the respiratory signal whereas MB-EPI showed reduced correlation with the cardiac signal and with cardiac spectral energy. The five correction schemes incrementally reduced physiological noise for both MB-EPI and regular EPI to some extent. Notably, the reduction appeared to be more pronounced for MB-EPI, with substantial trends observed over four of six metrics: correlation with respiratory, cardiac, and CRV signals, and cardiac spectral energy. The effect of the correction schemes on correlation with the CRV signal was particularly pronounced, especially when the CRV signal was included in the retrospective correction algorithm (Fig. 1d). In comparison, the correction schemes reduced physiological noise effects for regular EPI only in terms of the correlation with respiratory signal, and very slightly with the cardiac signal. Consequently, for the most aggressive physiological noise correction (RETROICOR+RVTcor+CRVcor), MB-EPI showed equivalent or better noise performance than regular EPI on four out of six metrics.Using various metrics, we have demonstrated that physiological noise characteristics are different in rs-fMRI data acquired by MB-EPI compared to those acquired by regular EPI. Complex effects were observed, with MB-EPI exhibiting worse physiological noise in comparison to regular EPI on three metrics, whereas equivocal or opposite results were observed on the other three. The mechanisms underlying these observations are unclear at present, and further investigation is required. It is possible that the slice-group effects in MB-EPI can exacerbate certain physiological noise signals, whereas aliasing can cause a noise cancellation effect in others. Interestingly, MB-EPI data appeared more amenable to physiological noise correction than regular EPI, indicating that comprehensive physiological noise correction may be more important for MB-EPI than for regular EPI data.