Slice window dynamic functional connectivity

Hello,

I’m currently working on dynamic functional connectivity analysis using movie-watching fMRI data acquired at 7T, comparing patients with bipolar disorder to healthy controls. My focus is on tracking connectivity changes within the limbic system throughout the movie, and relating these changes to emotion-related annotations from the stimulus.

I’m using Python 3 and Nilearn for the analysis.
The movie duration is approximately 8 minutes, and I have data from 20 subjects.

I’ve come across different recommendations in the literature for sliding window parameters in dFC analysis. Some studies use a window size ranging from 120 seconds to 15s, with various step sizes.

So far, I’ve been using:
window_size = 30 # corresponds to ~42 seconds with TR = 1.4s
step_size = 5 # ~7 seconds step

Knowing that :
Mean variation time (positive): 19.71 s
Mean variation time (negative): 14.55 s

Given the short duration of the movie (8 minutes), I’d like to ensure my parameters balance temporal resolution and stability of correlation estimates.

Do you have experience or recommendations regarding optimal window and step size for this kind of task? Is this more a software-related constraint, or mainly driven by neurophysiological considerations** and statistical trade-offs?

Thanks in advance !

Hi, Alienor
I am currently researching dynamic functional connectivity based on healthy individuals and schizophrenia patients. There is no definitive conclusion regarding the optimal length and step size of sliding time windows, though some studies suggest 30-60 seconds as a reasonable range. Emerging work has also attempted to address this issue. In my research, I typically use tapered windows with a step size of 1 TR and window lengths within 30-60 seconds. Finally, our latest review paper “A survey of brain functional network extraction methods using fMRI data” might be helpful to you. Please feel free to discuss any questions you may have.

best,
Songke Fang.