[Eeglablist] Multivariate Interaction Measure (MIM): ROIconnect

[Cedric Cannard]

Hi Rob,

My 2 cents would be that measures like GC or DTF can provide directionality and get closer to effective connectivity, but they are often more sensitive to noise, require stronger assumptions, and are less robust to source leakage.

But that would be a question for Pelligrini et al., I don't know the deep details on this. Maybe you can find some answers in these papers from the authors/developers/experts:
- https://urldefense.com/v3/__https://www.sciencedirect.com/science/article/pii/S1053811923003695?via*3Dihub__;JQ!!Mih3wA!EweMd1JxxRZPmbpcJyptNFyDQqQpJLV4tteYktQEESyQEWfoIjMy4MBfXtAIiCjgmkhLeifCxJqkwbB6kCbeD1A-pg$ 
- https://urldefense.com/v3/__https://www.biorxiv.org/content/10.1101/2023.10.26.564193v1__;!!Mih3wA!EweMd1JxxRZPmbpcJyptNFyDQqQpJLV4tteYktQEESyQEWfoIjMy4MBfXtAIiCjgmkhLeifCxJqkwbB6kCaKRBKJMQ$ 

As shown in Pellegrini et al. (2023), MIM performed among the best in simulations with known ground truth for detecting undirected phase-to-phase connectivity under realistic EEG conditions. It was compared with metrics like time-reversed Granger causality (TRGC), and stood out for accurately identifying true connections without inflating false positives due to source leakage or noise—issues that often affect directed metrics in EEG.

MIM is also multivariate, meaning it captures the full dimensionality of ROI activity instead of reducing it to a single time series (e.g., via PCA), which helps retain more physiologically meaningful information. In contrast, many directional methods require signal simplification and are more sensitive to the chosen aggregation method.

So while MIM doesn’t provide directionality, its strength lies in mapping reliable undirected functional interactions, which is often a necessary first step before applying more assumption-heavy directional metrics. In practice, I see value in using MIM to define a robust FC structure, and then exploring directionality on those links if needed.

If directionality is more important than identifying reliable functional networks, you can switch the algorithm to TRGC in the functions input in roiconnect. If I recall like this:

pop_roiconnect(EEG, 'connmethod', 'trgc');

or 

roiconnect(EEG, 'method', 'trgc');


Or, you could also try Tim Mullen's SIFT EEGLAB plugin: https://urldefense.com/v3/__https://github.com/sccn/SIFT__;!!Mih3wA!EweMd1JxxRZPmbpcJyptNFyDQqQpJLV4tteYktQEESyQEWfoIjMy4MBfXtAIiCjgmkhLeifCxJqkwbB6kCYKJ9dikQ$ 
It estimates effective connectivity (directional and causal interactions) using multivariate autoregressive (MVAR) models to compute metrics like Granger causality, DTF, and PDC. 

Hope this helps answer your concerns,

Cedric



On Tuesday, 8 July 2025 at 07:52, Rob Coben <drcoben at gmail.com> wrote:

> Thanks for the info. Not sure the value in this when there are other metrics that show directionality and causality getting one closer to effective connectivity.
> 
> Any info on this would be appreciated.
> 
> Rob
> 
> > On Jul 7, 2025, at 12:51 PM,
> > 
> > Cedric
> > 
> > 
> > Cannard via eeglablist eeglablist at sccn.ucsd.edu wrote:
> > 
> > Hi Tyson,
> > 
> > My understanding is that the MIM used in ROIconnect is designed to quantify the strength and consistency of phase-to-phase coupling between multivariate signals from different brain regions. It ranges from 0 to 1, with higher values indicating stronger consistent coupling, but it is insensitive to the sign of the phase difference (that is, it does not differentiate between in-phase and anti-phase relationships). Instead, it captures the magnitude of linear dependencies between multivariate time series across regions, regardless of their polarity. Not all FC metrics are constrained to [0, 1]. For instance, Pearson correlation ranges from -1 to 1, reflecting both direction and strength of linear association. However, many coherence-based and information-theoretic FC measures, including MIM, are non-negative and often bounded between 0 and 1, because they assess magnitude or shared variance/information, rather than direction or sign.
> > 
> > In short, MIM measures how strongly, not how directionally, the multivariate phase dynamics of two regions are related. It's a robust choice for undirected FC in the presence of source mixing and is not meant to infer polarity (like in-phase vs. anti-phase), only consistency of interaction.
> > 
> > Cedric
> > 
> > On Sunday, 6 July 2025 at 10:08, Dr Tyson Perez DC via eeglablist eeglablist at sccn.ucsd.edu wrote:
> > 
> > > Hi,
> > > 
> > > When I create my functional connectivity/correlation matrices using
> > > ROIconnect, MIM values range from 0 to 1 (i.e., no negative values). Is the
> > > MIM essentially computing the *consistency *of phase relationship between
> > > brain regions irrespective of whether they are in-phase or out-of-phase?
> > > 
> > > Cheers,
> > > Tyson Perez, DC, PhD
> > > _______________________________________________
> > > To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu or visit https://sccn.ucsd.edu/mailman/listinfo/eeglablist  .
> > > _______________________________________________
> > > To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu or visit https://sccn.ucsd.edu/mailman/listinfo/eeglablist .