Abstracts

Rationale: The human auditory system is optimized to identify minute spectral and spatial changes in time-based signals. However, traditional assessment methods for characterizing and classifying different types of epileptiform activity rely on visual displays that do not readily reveal these signal changes, thus requiring years of training that is typically reserved for medical personnel. A sound-based method for EEG classification may be used as an adjunct to current analysis methods and may improve the identification rate of spectral and spatial shifts in epileptiform activity as compared to visual display methods, in addition to requiring less training for proper onset identification. Methods: We developed a new method that directly converts EEG data into musical sound without relying on an intermediate classification stage, thus offloading the signal analysis to the listener's auditory system. The present sonification algorithm uses a bank of stored short musical sounds with varying musical pitch and sound qualities that are each assigned to one channel of an EEG recording. The algorithm imprints the spectrum of any given EEG channel onto a unique musical sound, thus revealing the shifts in spectral quality of the signal over time. In addition, each EEG-sound channel is assigned to a specific location in the stereo sound field, thus capturing the travel of inter-hemispheric seizure activity. Results: We applied this method to 20 samples of EEG data (mean 1 minute 43 seconds, standard deviation 1 minute 25 seconds)  recorded during routine evaluation of 4 patients undergoing monitoring for epileptic activity. Each sample represented a mix of non-seizure and seizure activity. A senior epileptologist identified onsets of epileptiform activity and shifts in type of activity within each sample. Our preliminary tests with three non-medical personnel showed that onset and change in type of epileptiform activity were captured using this sonification method. Multiple onsets of changing EEG activity were correctly identified in the sonified EEG samples. Participants noted that seizure activity sounded significantly different than non-seizure activity and that little training was required for identification. Spiking activity was particularly apparent and well-localized in the audio samples. Conclusions: We successfully implemented a novel EEG sonification algorithm designed to capture spectral and spatial shifts in ongoing epileptiform activity. Future validation steps will compare large cohorts of medical and non-medical personnel for identification accuracy for several classes of seizure activity. Funding: Not applicable