Authors:

TAYLOR, J.R. & HENSON, R.A.

Reference:

Year of publication:

2009

CBU number:

6986

Abstract:

Population inference from Magnetoencephalographic (MEG) data is complicated by MEG\'s sensitivity to neuroanatomical variability between individuals and to differences in head position relative to the sensors. Further, existing approaches rarely compensate for multiple comparisons over space and time. In the present study, these problems were overcome by (i) virtually transforming all subjects\' heads to a common location, (ii) applying random field theory to statistical parametric maps (SPMs), and (iii) constraining distributed source dipoles to a canonical cortical mesh, inverse-normalised to each subject\'s native space. Nineteen right-handed English adults saw words (W) and pronounceable pseudowords (PW; 480 each; duration 300 ms; presented sequentially at fixation) and indicated whether each was a word or nonsense word (button-press; hand counterbalanced). Whole-head MEG was acquired with 306 sensors (Elekta Neuromag). Neuromag\'s MaxFilter utility was used to apply signal-space separation to remove noise, for head-movement compensation, and for head-position transformation. Blink artefacts were removed using independent component analysis (EEGLAB). All further processing was done with SPM/Matlab. 3D SPMs on sensor data (2D topography x time) revealed significant W/PW differences from 300-400ms and 525-675ms, with topographies suggesting dipolar sources in bilateral temporal (left>right) and parietal regions. For each subject, source localisation was performed on epochs (-300 to 1000 ms; several approaches were evaluated). Source estimates were converted to image volumes for each condition in time-windows identified by sensor SPMs, smoothed (12-mm Gaussian kernel), and normalised. Group SPMs revealed activity (pseudowords>words) in perisylvian language regions; activity showing the opposite pattern (words>pseudowords) was localised to right parietal cortex.