The effect of reference choices on the spatio-temporal analysis of brain evoked potentials: the use of infinite...

Reference is a very virtual issue in EEG and ERP. Understanding the difference of various references will make the applications more confident. In this work, somatosensory evoked potential (SEP) with stimulation on the right hand was studied. The SEP spatio-temporal analysis was conducted comparatively on six references, left mastoid (contralateral mastoid reference, CM), right mastoid (ipsilateral mastoid reference, IM), linked mastoids (LM), average reference (AR), vertex reference (Cz) and the infinity reference (IR) newly proposed in 2001. Among the six, CM is the one used in actual recordings, and the other five are obtained by off-line re-referencing. The comparison is conducted on four selected components (P30 ms, P40 ms, N90 ms and P230 ms) in both temporal and spatial aspects. The results show that references may have a distinct influence on the amplitudes of the scalp potentials, with relative error at some electrodes larger than 500%, and for some electrodes it may even change the polarity. Pair-wise multiple comparison (Tukey test) shows that the differences of peak values among various references are very significant (P<0.001) between Cz and IR\CM\IM\LM, and significant (P<0.01) between Cz and AR for component N90 ms; very significant (P<0.001) between Cz and IR\CM\IM\LM\AR, significant between IMLM and AR (P<0.01), CM and AR (P<0.05) for component P230 ms. The amplitude value order is CM/IM> or =LM>IR>AR>Cz. The two-ways (the six references vs. the four Peaks) repeated measures ANOVA test shows the effect of different references depends on various components; there is a statistically significant interaction between reference and the peak (P=<0.001). While for the spatial map of the potential amplitude, references will not affect the amplitude map shape if the color-bar is selected automatically, but if a fixed color-bar is chosen for data of various references, they may show some differences. These results mean a common reference is important for producing a comparable result between labs. As IR is theoretically a constant reference, we recommend it as the common choice in the future.