A correlation study of averaged and single trial MEG signals: the average describes multiple histories each in a different set of single trials

Our understanding of the link between electrical events in the brain and behaviour is based on indirect measures. Positron Emission Tomography (PET) and functional Magnetic Resonance Imaging (fMRI) rely on haemodynamic processes which are slower by two to three orders of magnitude than the processes characterizing normal and pathological brain function. Direct invasive measurements of the electrical activity on the other hand produce too local a view which fails to show the large scale coherence which sustains awareness and cognition. On the opposite extreme, gross measures of the electrical activity like Electroencephalography (EEG) and single or few channel Magetoencephalography (MEG) had until recently to rely on simplistic point like models extracted from the averages of many repetitions of physiologically irrelevant stimuli. The introduction of multichannel probes with over 30 channels (Hämälainen et al. 1993), and the use of distributed source analysis (Ioannides et al. 1990a) opened up for the first time the possibility to study the response of single trials. In this work we address directly the question how representative is the description of events extracted from the analysis of the average signal. We use the simplest possible example: the cortical response to a simple 1 kHz tone, focusing on the early and by general admission "automatic" response around 100 ms after stimulus onset. To avoid the confounding inter-subject variability we have studied the responses over the left and right cortical areas to ipsi- and contralateral stimulation in a single subject; for testing reproducibility, we have used both the eyes open and eyes closed conditions. Since the computational demands involved in extracting a full three dimensional description from each trial are too great, we have complemented the distributed source analysis with special techniques, which allow us to scan through each and every single trial and identify each cortical activation similar to the ones picked out in the average signal. We are thus able to show conclusively that the sequence of events suggested by the analysis of the average signal is not representative of what is happening in individual trials. The sequence is made up of events which occurred in different trials reflecting probably the existence of many parallel routes each of which leads from the input at the ear to a final "computation".