Decoding dynamic brain patterns from evoked responses: A tutorial on multivariate pattern analysis applied to time-series neuroimaging data

摘要

Multivariate pattern analysis (MVPA) or brain decoding methods have becomestandard practice in analysing fMRI data. Although decoding methods have beenextensively applied in Brain Computing Interfaces (BCI), these methods haveonly recently been applied to time-series neuroimaging data such as MEG and EEGto address experimental questions in Cognitive Neuroscience. In atutorial-style review, we describe a broad set of options to inform futuretime-series decoding studies from a Cognitive Neuroscience perspective. Usingexample MEG data, we illustrate the effects that different options in thedecoding analysis pipeline can have on experimental results where the aim is to'decode' different perceptual stimuli or cognitive states over time fromdynamic brain activation patterns. We show that decisions made at bothpreprocessing (e.g., dimensionality reduction, subsampling, trial averaging)and decoding (e.g., classifier selection, cross-validation design) stages ofthe analysis can significantly affect the results. In addition to standarddecoding, we describe extensions to MVPA for time-varying neuroimaging dataincluding representational similarity analysis, temporal generalisation, andthe interpretation of classifier weight maps. Finally, we outline importantcaveats in the design and interpretation of time-series decoding experiments.