Abstract
Voxelwise encoding models of BOLD signals offer insight into how information in visual scenes is simultaneously represented in visual cortex. However, a complete understanding of how the brain internalizes visual information requires an understanding of the different transformational states of that information over time. Electroencephalography (EEG) has become a popular technique to understand the nature of those states, but suffers from dipole cancellation, thereby precluding a spatially complete signal of scene information. To circumvent that problem, we present the dynamic electrode-to-image (DETI) mapping procedure. The DETI procedure is an encoder-based approach that capitalizes on the state-space geometry of images and visual evoked potentials (VEPs) to map VEP signals to each location within scene images. Specifically, DETI mapping reduces the dimensionality of VEP signals from different electrodes over time and then maps those signals via a log-Gabor encoding model to each pixel within specific images. We applied this method to data gathered in a standard VEP paradigm whereby participants (n = 24) viewed 80 grayscale scene images (19.5 degrees of visual angle) while undergoing 128-channel EEG. DETI mapping revealed an interesting, possibly two-stage, pattern of transformational states that begin with a low spatial frequency (LSF) state (~50 ms), followed by a high spatial frequency (HSF) state (~70 ms to ~140 ms). Starting around 150 ms, the image transformations undergo what appears to be intermittent LSF transformations at ~180 ms and ~260 ms, possibly indicative of recurrent processes. Further, time-time regression analyses show that local regions within scenes undergo relatively unique spatiotemporal transformations over time, suggesting different temporal stages of local prioritization of scene information. The DETI mapping procedure therefore holds much potential to better understand the spatiotemporal states of visual information, thereby offering insight into how the early visual code shapes and refines higher semantic representations of our visual world.