Electric potentials and magnetic fields generated by ensembles of synchronously active neurons in response to external stimuli provide information essential to understanding the processes underlying cognitive and sensorimotor activity. Interpreting recordings of these potentials and fields is difficult as each detector records signals simultaneously generated by various regions throughout the brain. We introduce the differentially Variable Component Analysis (dVCA) algorithm, which relies on trial-to-trial variability in response amplitude and latency to identify multiple components. Using simulations we evaluate the importance of response variability to component identification, the robustness of dVCA to noise, and its ability to characterize single-trial data. Finally, we evaluate the technique using visually evoked field potentials recorded at incremental depths across the layers of cortical area VI, in an awake, behaving macaque monkey.
2
INTRODUCTIONThe field of neuroelectrophysiology relies on the analysis of electric potentials or magnetic fields produced by the brain in response to sensory stimulation, or in association with its cognitive and/or motor operations. These signals arise from transmembrane current flow produced by multiple ensembles of synchronously firing neurons. Far from being independent, these neural ensembles, also referred to as generators or sources, are often dynamically coupled in unknown ways that are of interest to the experimenter. Unfortunately, recording channels such as electrodes in electroencephalography (EEG) and superconducting quantum interference devices (SQUIDS) in magnetoencephalography (MEG) record linear mixtures of the activity from these sources in addition to ongoing background activity and sensor noise. Thus, the individual responses of each source are mixed within the recorded signal making it difficult to identify them and study their dynamical interactions. Furthermore, it is standard practice to enhance the signal-to-noise ratio by averaging event-related potentials (ERPs) over a number of experimental trials. However, implicit in this construction is the assumption that the evoked waveform is constant over trials and that any variability represents noise. In this practice, the possibility of assessing trialdependent effects in the data is sacrificed.The last decade has seen great developments in linear blind source separation (BSS) and independent component analysis (ICA) techniques, such as Infomax ICA (Bell & Sejnowski, 1995), FastICA (Hyvarinen & Oja, 1997), and second-order blind identification (SOBI) (Belouchrani et al., 1993). These algorithms have been useful in identifying sources in EEG and MEG signals using both ensemble-averaged data (Makeig et al., 1997;Sarela et al., 1998; Vighrio et al., 1999) and single trials (Jung et al., 1999;Cao et al., 2000;Makeig et al., 2002;Tang et al., 2002). However, with the exception of SOBI, the general assumption that the ERP sources are independent is physiologically implausible. It is hard to argue that acti...