The present study clarified the swallowing-related neural activities using human intracranial electrodes. Eight epileptic participants fitted with intracranial electrodes on the orofacial cortex were asked to swallow a water bolus, and cortical oscillatory changes were investigated. High γ (75-150 Hz) power increases associated with swallowing were observed in the subcentral area. To decode swallowing intention, ECoG signals were converted into images whose vertical axes were the electrode’s contacts and whose horizontal axis was the time in milliseconds; these findings were used as training data. Deep transfer learning was carried out using AlexNet, and the power in the high-γ band was used to create the training image set. The accuracy reached 74%, and the sensitivity reached 83%. We showed that a version of AlexNet pre-trained with visually meaningful images can be used for transfer learning of visually meaningless images made up of ECoG signals. This study demonstrated that swallowing-related high γ activities were observed in the subcentral area, and deep transfer learning using high γ activities enabled us to decode the swallowing-related neural activities.
