Localization of individual area neuronal activity N. Hironaga ⁎ and A.A. Ioannides Laboratory for Human Brain Dynamics, RIKEN Brain Science Institute (BSI), Wako-shi, Saitama 351-0198, Japan Kyushu Institute of Technology, Department of Brain Science and Engineering, Wakamatsu Ku, 2-4 Hibikino, Kitakyushu, 808-0196, Japan Received 21 November 2005; revised 26 September 2006; accepted 17 October 2006 Available online 21 December 2006 A family of methods, collectively known as independent component analysis (ICA), has recently been added to the array of methods designed to decompose a multi-channel signal into components. ICA methods have been applied to raw magnetoencephalography (MEG) and electroencephalography (EEG) signals to remove artifacts, especially when sources such as power line or cardiac activity generate strong components that dominate the signal. More recently, successful ICA extraction of stimulus-evoked responses has been reported from single-trial raw MEG and EEG signals. The extraction of weak components has often been erratic, depending on which ICA method is employed and even on what parameters are used. In this work, we show that if the emphasis is placed on individual “independent components,” as is usually the case with standard ICA applications, differences in the results obtained for different components are exaggerated. We propose instead the reconstruction of regional brain activations by combining tomographic estimates of individual inde- pendent components that have been selected by appropriate spatial and temporal criteria. Such localization of individual area neuronal activity (LIANA) allows reliable semi-automatic extraction of single-trial regional activations from raw MEG data. We demonstrate the new method with three different ICA algorithms applied to both computer- generated signals and real data. We show that LIANA provides almost identical results with each ICA method despite the fact that each method yields different individual components. © 2006 Elsevier Inc. All rights reserved. Keywords: MEG; ICA; Regional activation; Single-trial analysis Introduction Raw magnetoencephalography (MEG) and electroencephalo- graphy (EEG) signals are contaminated by interference from ambient noise and unrelated signals generated by the subject. Although we are not interested in these noise components, it is important to understand their nature so that they can more effectively be identified and eliminated. Ambient noise includes interference from commercial power lines (operating at 50 Hz in our laboratory), its harmonics, and DC drift, as well as electronic noise from thermal sensor instabilities and the acquisition system. Other unwanted components include strong signals from the heart, facial muscles, eye activity, and so on. In most cases, we are interested only in part of the brain activity—very often a small fraction of what goes on. If a convenient time reference can be defined—for example the onset of repeated stimulation—then averaging can be used to capture brain activity that is time-locked to the stimulus onset. Since the mid-1990s, independent component analysis (ICA) has been applied in MEG/EEG studies too numerous to adequately summarize here. In brief, early applications focused on the identification and elimination of artifacts (Vigario, 1997; Barros et al., 1998; Jung et al., 2000a,b; Joyce et al., 2004; Barbati et al., 2004; Wallstrom et al., 2004). Automatic identification of artifacts was suggested to be the main advantage of ICA-based methods compared to the more widely used principal component analysis (PCA). The application of ICA to event-related studies was initially introduced by the Salk Institute group (Makeig et al., 1997) and soon after by Vigario and colleagues (1998, 1999) in Finland using FastICA. Source localization of ICA components for medical application was initially proposed for epileptiform discharges (Kobayashi et al., 1999, 2002a,b; Ossadtchi et al., 2004). The Salk Institute group has reported ICA applications at both the average (Makeig et al., 1997, 1999) and single-trial level (Jung et al., 2001; Makeig et al., 2002; Delorme et al., 2002) using INFOMAX. Applying ICA to averaged data has also been discussed by the same group (Delorme and Makeig, 2004). Tang and colleagues (2002a,b, 2005) reported an improvement in response onset detection in single trials using second-order blind identification (SOBI). Lee and colleagues (2003) focused on the relationship between responses evoked by stimuli and that of the ongoing background activity using FastICA. MEG and EEG are the only techniques offering non-invasive measures of macroscopic brain activity with millisecond temporal resolution. In the case of MEG, strong claims have also been made for high spatial resolution (Moradi et al., 2003), but the prevailing view is that the spatial resolution of MEG is not as good as that of www.elsevier.com/locate/ynimg NeuroImage 34 (2007) 1519 – 1534 ⁎ Corresponding author. Laboratory for Human Brain Dynamics, RIKEN Brain Science Institute (BSI), 2-1 Hirosawa, Wako-shi, Saitama 351-0198, Japan. Fax: +81 48 467 9731. E-mail address: hironaga@brain.riken.go.jp (N. Hironaga). Available online on ScienceDirect (www.sciencedirect.com). 1053-8119/$ - see front matter © 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.neuroimage.2006.10.030