2020/01/10 by Jeff Hanna, Hanna, Jeff, Cora Kim +3
Computer Science · Neuroscience · #Blind Source Separation Techniques #EEG and Brain-Computer Interfaces #Neural dynamics and brain function
paper · pdf · doi:10.48550/arxiv.2001.03397
Background: Many magnetoencephalographs (MEG) contain, in addition to data\nchannels, a set of reference channels positioned relatively far from the head\nthat provide information on magnetic fields not originating from the brain.\nThis information is used to subtract sources of non-neural origin, with either\ngeometrical or least mean squares (LMS) methods. LMS methods in particular tend\nto be biased toward more constant noise sources and are often unable to remove\nintermittent noise.\n New Method: To better identify and eliminate external magnetic noise, we\npropose performing ICA directly on the MEG reference channels. This in most\ncases produces several components which are clear summaries of external noise\nsources with distinct spatio-temporal patterns. We present two algorithms for\nidentifying and removing such noise components from the data which can in many\ncases significantly improve data quality.\n Results: We performed simulations using forward models that contained both\nbrain sources and external noise sources. First, traditional LMS-based methods\nwere applied. While this removed a large amount of noise, a significant portion\nstill remained. In many cases, this portion could be removed using the proposed\ntechnique, with little to no false positives.\n Comparison with existing method(s): The proposed method removes significant\namounts of noise to which existing LMS-based methods tend to be insensitive.\n Conclusions: The proposed method complements and extends traditional\nreference based noise correction with little extra computational cost and low\nchances of false positives. Any MEG system with reference channels could profit\nfrom its use, particularly in labs with intermittent noise sources.\n