Poster Presentation

Poster A in Poster Session A: Tuesday, August 4, 9:30 – 11:15 am, Kimmel Center, Shorin & Rosenthal Rooms

RSVP-induced ERP Class Posteriors Reflect Graded Neural Perception of Humanoid Images

Basak Celik¹, Catherine Huang², Cindee Madison¹, Santosh Mathan², Bryant Gipson², Andrew J Smart², Deniz Erdogmus¹; ¹Northeastern University, ²Google

Presenter: Basak Celik

In the rapid serial visual presentation (RSVP) paradigm, EEG event-related potentials (ERPs) are captured when a target stimulus appears in the stimulus sequence. In the context of detecting target objects (e.g. humans) in images, it is uncertain whether an ERP will manifest in EEG readings when a sequence includes a small ratio of pseudotargets (e.g. non-human but human-shaped humanoids) interspersed with actual human targets. We studied this question through a binary EEG classification model that yields target perception scores for human presence in images. This model produces a value in [0,1] depending on image content, from EEG time-locked to image onset in RSVP. The model was calibrated using a set of EEG responses to a combination of distractor (nontarget) and target images shown to multiple subjects within the RSVP paradigm. Target class posterior given EEG temporally aligned to image onset was used as a target perception score. Analyses on test imagery demonstrated that, as anticipated, distractor images result in EEG signals that tend to receive lower target perception scores than that of target images. More interestingly, pseudotarget (humanoid) test images that were intentionally designed to induce confusion received slightly lower target perception scores than target images. While the analyses and observations are based on data from limited number of subjects and trials, the results are promising as they indicate degraded target recognition response to humanoid images in contrast to genuine targets.

Topic Area: Methods, Tools, Theory & Neural Coding