In clinical practice, EEGs are assessed visually and recordings with reduced number of electrodes or artefacts make the assessment more difficult. Present techniques for upsampling or restoring channels utilize different interpolation strategies by taking averages of neighboring electrodes or fitting surfaces to the electrodes. These techniques usually perform better for higher electrode densities and values interpolated at areas far from electrodes can be unreliable. Using a method that instead learns the statistical distribution of the cortical electrical fields and predicts values of missing electrodes may yield better results. Generative networks based on convolutional layers were trained to upsample from 4 or 14 channels or restore single missing channels to recreate 21 channel EEGs. Roughly 5,144 hours of data from 1,385 subjects of the Temple University Hospital EEG data Corpus were used for training and evaluating the networks. The results were compared to interpolation by spherical splines and a visual evaluation by board certified clinical neurophysiologists was conducted. In addition, the effect on performance due to the number of subjects used for training was evaluated. The generative networks performed significantly better overall compared to spherical spline interpolation. There was no difference between real and network generated data in the number of examples assessed as fake by experienced EEG interpreters. On the contrary, the number was significantly higher for data generated by interpolation. Network performance improved with increasing number of included subjects, with the greatest effect seen for 100. Using neural networks to restore or upsample EEG signals is a viable alternative to existing methods.### Competing Interest StatementThe authors have declared no competing interest.

Virtual EEG-electrodes: Convolutional neural networks as a method for upsampling or restoring channels