Introduction

Mapping and characterizing the human brain networks using resting-state fMRI (rs-fMRI) has gained substantial attention in recent years ¹. rs-fMRI is conventionally performed using a single-band single-echo EPI sequence, which is limited by low temporal and spatial resolutions, susceptible to signal dropouts in the vicinity of air-tissue interfaces and vulnerable to superficial correlations owing to noise sources not related to neuronal activities in BOLD signal ². In the present study, we estimated dynamic functional connectivity (FC) using a multi-band multi-echo EPI (M2-EPI) imaging method ³, which has higher temporal resolution, less susceptibility related signal dropouts and better signal-to-noise ratio.

Methods

Seventeen participants (age, 42.53 ± 16.85 years; range 20-68 years; 7 females) were included in the study. Resting-state functional images were acquired using a M2-EPI pulse sequence, which combines multiband excitation and multi-echo EPI readouts, with the following parameters: TR = 840 ms, TE = 12/30/48 ms, FOV = 220 x 220 mm², matrix = 68 x 68, slice thickness/gap = 3.4/0.4 mm, 33 slices, multi-band factor = 3, GRAPPA factor = 2. The M2-EPI scan consisted of 980 contiguous image volumes (823.2 s). A single-shot EPI (SS-EPI) imaging (TR = 2000 ms, TE = 30 ms, FOV = 220 x 220 mm², matrix = 64 x 64, slice thickness/gap = 3.4/0.4 mm, 33 slices) with 400 image volumes (800 s) was performed for comparison. ME-ICA ⁴ was applied to M2-EPI data for multi-echo combination and denoising. Ninety ROIs were obtained according to the automated anatomical labeling atlas. A sliding window of 100 s was used to estimate dynamic FC between ROIs. Dynamic FC was further converted to Fisher Z value. Temporal variation of dynamic FC was calculated by dividing the SD by the mean across time windows. Statistical comparison between temporal variation of dynamic FC obtained from M2-EPI and SS-EPI data was performed using paired t-test. rs-fMRI data was also used to classify the brain activity into different states, termed FC states. To obtain FC states, k-means clustering was applied to correlation matrices with L1 distance. The number of cluster was selected by using the elbow criterion of silhouette values.

Results

Fig 1 displays correlation matrices within different time windows in a representative participant, which shows variation of FC over time. Lower temporal variation of dynamic FC was observed using M2-EPI relative to SS-EPI (paired t-test, P < 0.001) (Fig 2). FC states derived from k-mean clustering algorithm with k = 9 were shown in Fig 3. Each matrix represents the cluster centroid, indicating a state of dynamic FC. Robust differences among FC states could be observed through visual inspection. Fig 4 further presents these differences with weighted graphs constructed from correlation matrices in Fig 3. The components of default mode network were present in State 5, which occurred in 27.09% of the time. In State 8, the occipital and parietal regions were densely connected and formed a community. Transitions between different FC states as a function of time were found during resting state (Fig 5).

Discussion and Conclusion

Dynamic FC and its states can be observed using M2-EPI rs-fMRI data. Compared to conventional SS-EPI imaging, reduced temporal variation in dynamic FC is achieved using M2-EPI data acquisition, possibly due to fast acquisition and improvement in noise reduction. The use of M2-EPI BOLD imaging may help to better delineate the architecture of resting-state functional networks.

Acknowledgements

No acknowledgement found.