Graph analysis of resting state eeg data using MNE and Networkx

Preprocessing

• Import the data, filter them (the mne filters already use zero-phase filters) at 1-30Hz.
• Set an average reference.
• Check and exclude bad electrodes before doing ICA.
• Do a first visual inspection of raw data, excluding segments containing obvious artifacts.
• Calculate the rejection thresholds that I will pass to ICA
• Run ICA using the extended-infomax method
• Visual inspect ICA components to check for the ones representing eye-movements or blinks.
• Run an automatic procedure that should highligh those components
• Apply ICA
• Create the epochs and perform a last visual inspection to exclude bad epochs.
• Save the epochs Connectivity matrix
• Connectivity matrix calculation using the PLI method.
• Create a graph using those values. Given that I am using a 128-electrodes system, I decided to remove connections of electrodes with a distance below 3cm. That's an arbitrary choice though.
• Plot the connectivity matrix (just for visualization)
• Derive the MST subnetwork (which is actually a maximum spanning tree). It has 124 nodes and 123 edges as I excluded 4 EOG electrodes.
• Plot the adjacency matrix and degree distribution
• Calculation of all the metrics used to characterize MSTs: degree, leaf nodes, leaf fraction , max degree, diameter, eccentricity, betweenness centrality, max betweenness centrality, tree hierarchy and degree correlation.

Steps for the one subject analysis 0) Global variables: info about the subject

1. Filtering, re-referencing and visual inspection
2. ICA
3. Epoching
4. Connectivity matrix
5. Plots
6. Metrics
7. Saving the results

• Graph analysis of functional brain networks: practical issues in translational neuroscience; Vellani et al. (2014)
• The trees and the forest: Characterization of complex brain networks with minimum spanning trees; Stam et al. (2014)