Abstracts

Rationale: Over the last several decades, a great deal of research has investigated the neural basis of emotion. While the advent of fMRI has led to an explosion of studies in humans, the literature on emotion utilizing direct recordings from the human brain remains limited, and much of it has focused on measuring broadband evoked responses to different facial expressions. Yet, linguistic stimuli have the potential to evoke emotional connotations with a degree of specificity and nuance difficult to capture with facial expressions, and specific neural systems are necessary to process these stimuli for their emotional content. Evoked high gamma band activity has emerged as a robust marker of localized task-related neural activity that is best measured with invasive electrodes. We measured high gamma activity in response to both linguistic and face stimuli and employed a machine learning algorithm to identify emotion-relevant information in the signal in multiple brain locations. Methods: Intracranial EEG was recorded from 1159 electrodes in 15 patients while they looked at visual stimuli containing positive and negative emotional as well as neutral content as conveyed via single printed words or images of facial expressions. High gamma activity (80-150 Hz) was extracted from the signal using the Filter-Hilbert method. Diaquadratic discriminant analysis was used to train a classifier to discriminate stimuli by emotional content on a single-trial basis. Electrodes were localized to specific brain regions based on the atlas by Destrieux using Freesurfer and iELVis software.  Results: The classifier analyses demonstrated that it is possible to distinguish emotionally negative from emotionally neutral stimuli, both when using words or faces as inputs. These decoding analyses provide a documentation of the spatiotemporal dynamics underlying emotional processing in cortical circuits. Furthermore, the neural structures subserving processing of emotional valence for highly distinct types of stimuli, words versus faces, were directly compared for the same subjects and electrodes.  Conclusions: Discriminant analysis can be used to with intracranial high gamma activity to decode emotional content from single trials. This technique reveals regionally and temporally-specific emotion-related information in the high gamma band signal. Funding: NIH 1R21MH107820-01A1