Decoding mental states in bistable perception by using source based wavelet features

This work presents an efficient application of machine learning and signal processing algorithms in decoding the mental states from brain signals. We decoded two internal mental states, transition and maintenance, representing the process of switching and maintaining a perception in bistable perception, respectively. The underlying sources were reconstructed from the Magnetoencephalogram (MEG) signals recorded from human participants while they were presented with six different conditions of bistable stimuli. We extracted features using complex Morlet wavelet transform that captured the temporal dynamics of large scale brain oscillations in the source domain. The mental states were predicted on a trial-by-trial basis using Support Vector Machine (SVM) and Artificial Neural Network (ANN) classifiers along with Fisher's ratio based feature selection technique. We achieved accuracies of 75.65% and 74.58% using SVM and ANN classifier, respectively. The analysis also exhibits the involvement of the sources of parietal, temporal and cerebellum areas in characterizing the two mental processes.