Decoding of object categories from brain signals using cross frequency coupling methods

The procedure by which information about conceptual category of visual objects is encoded in brain signals, is the topic of many studies in the field of visual system. A phenomenon that is recently mentioned as one of information encoding strategies in the brain is cross frequency coupling (CFC), means the interactions between different frequency bands in physiological signals such as EEG. This interaction can occur between various components of such signals including amplitude or power, phase and frequency. Different types of CFC have been reported in various cognitive tasks and within or between different brain regions. However the role of CFC in encoding of information about the category of visual objects is greatly unknown and here we attempted to investigate it. So in this paper, we used machine learning algorithms to find out whether CFC contains such information. To this end, we recorded EEG from 10 participants while they were observing stimuli from 12 different visual object categories. Then amplitude-amplitude coupling (AAC), phase-amplitude coupling (PAC) and phase-phase coupling (PPC) within each electrode were calculated for the recorded signals in order to use them as input to SVM classifier. The results show that phase-phase coupling can provide more information about the category of objects compared to other types of CFC, by classification performance of 92.33% against 70.28% and 60.52% for phase-amplitude and amplitude-amplitude coupling, respectively. In addition, this feature was more informative when occurred between alpha and gamma frequency bands. The performance of classification by means of CFC features was then compared with the result of classification by wavelet transform on the same data. We observed that PAC can improve the performance of categorical-based classification relative to wavelet coefficients (with the performance of 70.73%). So we can conclude that cross frequency coupling encompass useful information about semantic category of stimuli which is not available in time-frequency components of the signal obtained by wavelet transform.