Independent component analysis for unaveraged single-trial MEG data decomposition and single-dipole source localization

This paper presents a novel method for decomposing and localizing unaveraged single-trial magnetoencephalographic data based on the independent component analysis (ICA) approach associated with preand post-processing techniques. In the pre-processing stage, recorded single-trial raw data are 6rst decomposed into uncorrelated signals with the reduction of high-power additive noise. In the stage of source separation, the decorrelated source signals are further decomposed into independent source components. In the post-processing stage, we perform a source localization procedure to seek a single-dipole map of decomposed individual source components, e.g., evoked responses. The 6rst results of applying the proposed robust ICA approach to single-trial data with phantom and auditory evoked 6eld tasks indicate the following. (1) A source signal is successfully extracted from unaveraged single-trial phantom data. The accuracy of dipole estimation for the decomposed source is even better than that of taking the average of total trials. (2) Not only the behavior and location of individual neuronal sources can be obtained but also the activity strength (amplitude) of evoked responses corresponding to a stimulation trial can be obtained and visualized. Moreover, the dynamics of individual neuronal sources, such as the ∗Corresponding author. Present address: Department of Electronic Engineering, Saitama Institute of Technology, 1690 Fusaiji, Okabe, Saitama 369-0293, Japan. E-mail address: [email protected] (J. Cao). 0925-2312/02/$ see front matter c © 2002 Elsevier Science B.V. All rights reserved. PII: S0925 -2312(02)00519 -2 256 J. Cao et al. / Neurocomputing 49 (2002) 255–277 trial-by-trial variations of the amplitude and location, can be observed. c © 2002 Elsevier Science B.V. All rights reserved.