Effective connectivity and causality

The standard measure of functional connectivity, Pearson correlation between fMRI time series, also detects indirect connections: when two regions of the brain are not directly connected but connected by a chain of strongly correlated brain regions, their fMRI time series are strongly correlated though they are not physically connected. As a result, Pearson correlations are prone to detect false positive functional connections.
Several measures of effective functional connectivity, where indirect connectivity has been factored out, have been proposed to address this issue. The most common measure is the partial correlation, obtained by inverting Pearson correlations. I have recently investigated efficient strategies to compute partial correlations at full brain resolution, with the aim of generating better brain parcellations. Also, I showed that these methods could be used for analyzing the structural covariance matrices computed, at the level of a population, to reveal how brain regions observed via structural MRI are simultaneously affected by diseases, neurodevelopment, and aging.

The Matlab toolbox TORII provides a subset of the methods I have implemented to compute regularized inverse covariance matrices and partial correlations. download page

I recently realised that causality measures might constitute an interesting alternative to connectivity, because they offer a richer description (they providing a direction for each connection). I am currently exploring this alternative.


[A] Causality patterns extracted for the hundred unrelated HCP subjects. [B] A Mode of cortical thickness variations, obtained when inverting the associated covariance matrix.

associated publications