EEG channel interpolation using ellipsoid geodesic length

EEG channel interpolation is of great significance when the EEG signal from a channel is very low quality or missing altogether. This is particularly critical for low density EEG arrays employed in several clinical and research applications because the missing channel represents a large portion of the underlying cortical activity and adversely affects further data analysis and, potentially, diagnosis. For the same reasons, it is also critical for dry EEG applications when a disrupting contact is present. We present an approach for reconstructing a missing or poor EEG signal by combining signals from neighboring electrodes according to their distance from the electrode corresponding to the reconstructed signal. We used EEG data recorded from humans performing a cognitive task, omitted one channel at a time (to assess any spatial dependencies of our proposed approach), and reconstructed the omitted signal from the rest of the signals. Signals were reconstructed using Euclidean distance, great circle distance, and ellipsoid geodesic length and compared to reference (omitted) signals by means of inspection and normalized mean square error. Pilot results indicated that the ellipsoid geodesic length gave the best signal reconstruction as it provided event related potential and scalp map estimates closest to the reference and the smallest average (across all omitted channels) normalized mean square error (NMSE) values.