Functions for performing sensor and source-space decoding analyses on Fieldtrip-processed MEG data using a Support Vector Machine for two-class problems, implementing different spatiotemporal feature selection approaches and cross-validation schemes.
Work in progress: functions for performing Representational Similarity Analysis of spatiotemporally resolved MEG data.
Reliant on the Fieldtrip toolbox for preprocessing, source localization, plotting, and templates, and LibLinear for SVM decoding.
An example dataset and script for running some decoding analyses are included in the demo directory and described here.
Functions for preprocessing sensor-level MEG data, source-reconstructing MEG data with a beamformer, and preparing data for decoding (creating pseudo-trials, whitening, getting spatial clustering information for spatially-resolved MVPA).
Functions for decoding MEG data across time (time-resolved and temporal generalization) and space (searchlight, region-of-interest, source/channel selection, whole-brain), with kfold or hold-out cross-validation.
The main functions use the LibLinear SVM. Functions based on LibSVM and Matlab SVM are included in subfolders.
Non-parametric significance testing with different methods of correcting for multiple comparisons.
Functions for plotting decoding results over time and in space (sensor/source space) and as movies (over time/on a rotating template brain).
Functions for Representational Similarity Analysis (work in progress): create dissimilarity matrices, correlate them with models, calculate a noise ceiling and randomize the correlations for significance testing.