Decoding Imagined Musical Pitch from Human Scalp Electroencephalograms

Abstract

Brain-computer interfaces (BCIs) can restore the functions of communication and control in people with paralysis. In addition to the currently proven functions restored by BCIs, it would enrich life if one could regain a function of musical activity. However, it remains largely unknown whether it is feasible to decode imagined musical information directly from neural activity. Among various musical information, this study aimed to decode pitch information directly from scalp electroencephalography (EEG). Twenty healthy participants performed a task to imagine one of the seven musical pitches (C4 - B4) randomly. To find EEG features for pitch imagination, we took two approaches: exploring multi-band spectral power at individual channels (IC); and exploring power differences between bilaterally symmetric channels (DC). We classified these features into the seven pitch classes using various types of classifiers. The selected spectral power features revealed marked contrasts between left and right hemispheres, between low-, (<13 Hz) and high-frequency (> 13 Hz) bands, and between frontal and parietal areas. The best classification performance for seven pitches was obtained using the IC feature and SVM with the average accuracy of 35.68{+/-}7.47% (max. 50%) and the average information transfer rate (ITR) of 0.37{+/-}0.22 bits/sec. Yet, when we decoded a different number of classes (K = 2 [~] 6) by grouping adjacent pitches, ITR was similar across K as well as between IC and DC features, suggesting efficiency of DC features. This study would be the first to demonstrate the feasibility of decoding imagined musical pitch directly from human EEG.