When estimating functional connectivity from magnetoencephalographic (MEG) signals, regions of interest (ROIs) are typically defined based on standard anatomical parcellations, which are not informed by the limited spatial resolution of MEG. In this talk, we present an unsupervised clustering approach to divide and downsize the cortical source-space into ROIs based on the forward operator. By explicitly taking into account the MEG spatial resolution, the algorithm produces ROIs with a lower cross-talk compared to anatomical parcellations.

This presentation is part of Minisymposium “MS18 - Functional neuroimaging methods for experimental data”
organized by: Anna Maria Massone (CNR - SPIN) .

Authors:

Sara Sommariva (Aalto University)

Narayan Puthanmadam Subramaniyam (Aalto University)

Lauri Parkkonen (Aalto University)

Keywords:

clustering, functional connectivity, machine learning, magnetoencephalography