Significance tests for MEG response detection

Magnetoencephalography (MEG) is a non-invasive neurophysiological technique with high temporal resolution. Nevertheless, low signal to noise ratio may hamper its fullest capability. Many confidence tests already exist to detect strong responses for signals corrupted by noise, and we have explored their use with experimentally obtained MEG signals. We find that the tests demonstrating the most power are the F-test and Rayleigh's phase coherence test. Due to the strongly non-Gaussian nature of the MEG noise, from both neural and external perspective, a signal which is purely noise often fails the marginal tests by exceeding the number of false positive allowed. A variation of the tests is suggested that ensures the average false positive for a large number of responses, excited at frequencies different than the frequency of interest, is below any desired threshold. This is implemented for the F-test, Rayleigh's phase coherence test, and the union of the two