User-friendly open-source pipeline for anatomically precise analysis of single-trial M/EEG

PROJECT SUMMARY
Magneto- and electroencephalography (M/EEG) measure human brain activity non-invasively, using sensors
placed on or above the head, with millisecond accuracy, and have long been used to study neural and cognitive
processes. This project is to develop a user-friendly, open-source analysis pipeline to make recent innovations
in analysis techniques available to a broad audience. First, M/EEG has traditionally been analyzed primarily as
event-related activity (an experiment is conceptualized as a series of trials, each constituting a unique event; for
example, the response to a syllable da). Such experiments typically require many repetitions of identical trials to
separate signal from noise in the M/EEG response. More recent techniques allow conceptualizing experimental
time as continuous, with different aspects of stimuli that are distributed in time evoking different overlapping brain
responses (for example, the auditory cortex response continuously “tracks” the acoustic envelope of speech, or
the visual cortex tracks the amount of visual motion in a movie, etc.). In this time-continuous analysis, no
repeated trials are necessary, and the signal is instead estimated using detailed models of the stimuli. Second,
while brain activity is measured outside of the head, it has long been possible to estimate anatomically localized
sources of this activity in the brain, and recent advances have greatly improved this source estimation. These
two advances (single-trial continuous analysis and improved source localization) have only recently been
combined successfully, and no off-the-shelf software package allows researchers to readily apply this method to
existing or new datasets. This project will develop such a software tool with an easy-to-use pipeline for group
level analysis. The technique that this tool will make available is essential for analyzing data from experimental
designs with naturalistic stimuli (for example, participants watching a movie or listening to an audiobook). Such
experiments offer enhanced validity, and are rich and versatile: researchers can test many different hypotheses
about what neural processes are evoked by the stimuli. However, the traditional averaging techniques that are
widely available are not remotely suitable. The software tool developed here is ideally suited for analyzing such
datasets. It will be made widely accessible and user-friendly, to lower the threshold for non-neuroscientists to
test hypotheses about neural representations. For example, existing datasets of audiobook listening could be
used by psychologists, computer scientists and linguists to test hypotheses about neural representations of
speech.

Grant Number: 5R01MH132660-02
NIH Institute/Center: NIH
Principal Investigator: Christian Brodbeck