Abstract
In this technical note, we present the challenges that prevent us from directly comparing sequences with and without MB acceleration at the single subject level. Using fMRI data collected with MB1S2 (TR 2.45s), MB2S2 (TR 1.22s) and MB4S2 (TR 0.63s), we note the CNR differences in the images acquired with different sequences which leads to global mean scaling that render the direct comparison of parameter estimates meaningless. Directly comparing t-values of participants across different acquisition sequences is meaningless because of the difference in degrees of freedom (df) introduced by the higher number of volumes acquired at higher multiband. Z-transformation of the t-statics to correct for the difference in degree of freedoms suggests that sequences without MB outperform sequences with MB acceleration. However, this may be due to an excessive penalty caused by inappropriate df estimation. Thus with the current evidence presented in this and previous studies that tested the impact of MB on task related-statistics, the field lacks empirical evidence for the effects of MB on individual subject statistics. We discuss the possible alternatives such as use of Bayesian statistics.