Search Papers | Poster Sessions | All Posters
Poster C16 in Poster Session C - Friday, August 9, 2024, 11:15 am – 1:15 pm, Johnson Ice Rink
Image-computable encoding models of human broadband iEEG responses to natural images reveal time-dependent representations
Ghislain St-Yves1, Harvey Huang2, Zeeshan Qadir2, Morgan Montoya2, Gregory Worrell2, Kai Miller2, Kendrick Kay3, Dora Hermes2, Thomas Naselaris1; 1University of Minnesota, Department of Neuroscience, 2Mayo Clinic, Department of Physiology and Biomedical Engineering, 3University of Minnesota, Department of Radiology
Image-computable models designed for prediction of fMRI BOLD signals have been shown to generalize well to iEEG broadband for simple stimuli. We show that models with complex feature spaces (DNNs) that have been used to predict natural image responses in fMRI signals can also be trained to predict iEEG broadband responses. The high temporal sampling afforded by iEEG signal enable us to precisely locate the onset of activity and characterize the temporal evolution of representational tuning at various recording sites. We show that the temporal onset is strongly correlated with network tuning depth across all subjects. Furthermore, we show that tuning properties in several channels vary over time after onset, with retinotopic representations subsiding and feature tuning drifting toward more semantic-like representations.
Keywords: Vision Encoding model Human iEEG Time-dependent representations