The acquisition of resting-state BOLD images is typically used to investigate the functional connectivity between brain regions. A novel use of resting-state data involves estimation of entropy.¹ Entropy is a measure of the complexity of signals. Highly predictable and regular signals have low entropy, while random and unpredictable signals have high entropy. Multi-scale entropy (MSE) is a recent approach to quantifying this complexity in brain signals that estimates the entropy present in a signal by averaging components of the time-series.² At higher scales (i.e., the averaged signal of a greater number of time-points) the effects of high-frequency noise are reduced, resulting in improved estimation of the underlying entropy. MSE has been used in a number of studies, revealing a decrease in entropy associated with aging.^3,4

While BOLD-derived MSE is a potential biomarker of aging and related processes, it is not yet understood how imaging acquisition factors might impact its computation. The aim of this study was to investigate whether MSE is influenced by multi-band acquisition of BOLD images. Given that age related decline in MSE has already been established, two groups (older versus younger) were used to evaluate the sensitivity of different sequences to biologically driven changes in entropy. We expected that the shorter TR facilitated by multi-band acquisitions would increase the sensitivity to age related changes.

Data for this study were acquired from the Nathan Kline/Rockland Enhanced study. 60 participants were included in total, which included 30 younger participants (Mean age = 23.37, SE = 0.56) and 30 older participants (Mean age = 75.83, SE = 0.81). Four images were acquired for each participant, which included: (1) an MPRAGE for anatomical coregistration, (2) a standard T2* BOLD weighted image with no mulit-band acquisition, (3) a multi-band acquisition with 4 times acceleration and TR of 1400ms, and (4) a multi-band acquisition with 4 times acceleration and TR of 645ms. For the first analysis, each resting state acquisition was concatenated to approximately 5 mins (2500ms TR = 120 vols, 1400ms TR = 215 vols, 645ms TR = 466 vols). All analyses were then repeated with the first 120 volumes only to investigate the contribution of TR vs number of volumes acquired.

All images were pre-processed using the functional connectivity toolbox.⁵ Structural volumes were coregistered to functional volumes and segmented into tissue classes. Functional volumes were normalised to MNI space. Physiological noise was removed using the aCompCor approach.⁶ MSE was computed using Matlab 2015a at five scales (1-5). Mean MSE for the grey matter for each brain was extracted in MNI space using the Harvard-Oxford subcortical atlas. Repeated measures ANOVAs were used to analyses the influences on MSE. ROC analyses were used to investigate the optimal condition for classifying older and younger participants. Classification accuracy was determined using the area under the curve (AUC).

The first analysis investigated the effect of the different acquisitions across all five MSE scales for the 5 minute and 120 volume conditions. This three way interaction was statistically significant, F(8,472) = 47.59, p < .001, η_g² = 0.01. Overall, MSE decreased with increasing scale. In the 5 minute condition, 2500ms TR produced higher MSE, while the two multi-band acquisitions were comparable. This difference was not seen in the 120 volume condition. This interaction is shown if Figure 1. As shown in Figure 2, classification accuracy was generally comparable across TRs and scales. The most accurate classification was obtained using data from the 5 minute acquisition with TR = 1400ms at scale 4 (AUC = .73 [.60,.86]). As shown in Figure 3, the two multi-band acquisitions produced the most accurate classification, however a consistent pattern did not emerge. These data suggest that the use of multi-band imaging to acquire resting data influences the computation of MSE. The two multi-band sequences used here produced lower MSE estimates compared to non-multi-band imaging, however this was only observed for acquisition times of equal length. This suggests that it was the number of volumes, and not the multi-band acquisition (i.e., shorted TRs), that drove these differences. While the multi-band sequences tended to separate the older and younger groups better, this trend was not consistent across all scales.Overall, these results suggest that multi-band acquisition might facilitate the accurate measurement of brain entropy at rest. Given the differences between acquisition methods, researchers should be careful to ensure that any comparisons of MSE values are not confounded by differences in acquisition length. Further research is needed to identify the contribution of additional factors, such as movement and signal outliers to the computation of MSE.