Truncated RAP-MUSIC (TRAP-MUSIC) for MEG and EEG source localization

Abstract

Electrically active brain regions can be located applying MUltiple SIgnal Classification (MUSIC) on magneto- or electroencephalographic (MEG; EEG) data. We introduce a new MUSIC method, called truncated recursively-applied-and-projected MUSIC (TRAP-MUSIC). It corrects a hidden deficiency of the conventional RAP-MUSIC algorithm, which prevents estimation of the true number of brain-signal sources accurately. The correction is done by applying a sequential dimension reduction to the signal-subspace projection. We show that TRAP-MUSIC significantly improves the performance of MUSIC-type localization; in particular, it successfully and robustly locates active brain regions and estimates their number. We compare TRAP-MUSIC and RAP-MUSIC in simulations with varying key parameters, e.g., signal-to-noise ratio, correlation between source time-courses, and initial estimate for the dimension of the signal space. In addition, we validate TRAP-MUSIC with measured MEG data. We suggest that with the proposed TRAP-MUSIC method, MUSIC-type localization could become more reliable and suitable for various online and offline MEG and EEG applications.

Locating highly correlated sources from MEG with (recursive) (R)DS-MUSIC

“…In this study, we used the vector forms for the localizers, that is, the candidate source orientations were not pre-determined in the forward model [31,24]. However, all the presented localizers can be easily transformed into their scalar versions.…”

“…The source orientations can be determined, e.g., by solving a generalized eigenvalue problem related to Eq. 5 [24].…”

“…The iteration is continued until all source dipoles have been found, which is typically indicated by a significant drop of the maximum value of µ RDS k (p), or post hoc, afterñ recursions, by using the drop of the maximum value of µ RDS k (p) as a classifier for true vs. false sources [28,24]. The classification can be done either by setting a fixed threshold for the localizer value maximum [28,22]), or adaptively, by monitoring the evolution of the localizer value maximum as a function of recursion steps.…”

“…The classification can be done either by setting a fixed threshold for the localizer value maximum [28,22]), or adaptively, by monitoring the evolution of the localizer value maximum as a function of recursion steps. In general, the adaptive approach should be more robust [21,3,24].…”

“…1). Similar approach has been used previously in [24] and [27]. The MEG forward solution for the source candidates in the scanning grid was computed by a linearcollocation boundary-element method (BEM) with the isolatedsource approach in a 3-layer head model including boundary surfaces for the scalp, skull, and brain [17,34,35].…”