Ana Rozman¹ (), Abigail Flowers¹, Jenny Bosten¹; ¹University of Sussex

SSVEP signals are strongly influenced by low-level visual features, and it is challenging to use them to extract responses to higher level visual features. Aiming to improve on existing approaches to isolating high-level (e.g., object-specific) responses using SSVEP, we developed an enhanced implementation of image scrambling methods. We presented food texture stimuli modulating at 2 Hz in 4 conditions, and recorded 64-channel EEG while participants passively viewed trials of 30 seconds duration. Two of the stimulus conditions were based on existing methods for targeting object responses using SSVEP: Semantic wavelet-induced frequency tagging (SWIFT: Koenig-Robert & VanRullen, 2013, NeuroImage), and a progressive (sinusoidal) phase scramble. Two additional conditions controlled for modulation of low-level image elements within each approach, where two different phase-scrambled images or two different SWIFT-scrambled images were modulated. In current implementations of SWIFT or phase-scrambling, SSVEP signals are assumed to isolate high level responses if low-level image features are constant over the stimulus modulation. Our additional conditions account for remaining modulations of low-level features when we compare SSVEP signals for (i) original versus scrambled stimulus modulations and (ii) scrambled versus scrambled stimulus modulations. We make these comparisons by regressing (i) against (ii) across the 64 channels, and analysing the residuals, using baseline-corrected and summed SSVEP amplitudes in the frequency domain for the harmonic series of 2 Hz. Results show that differences between conditions in signals captured by residuals are more anterior than signals based on raw modulations (e.g., SSVEPs to SWIFT modulations alone). These findings suggest that our protocol can better isolate a signature of higher-level responses to passively viewed object textures than scrambling methods without the additional control. Our next aim is to adapt these methods to study higher-level color representations using SSVEP.

Acknowledgements: The work was funded by ERC grant 949242 COLOURCODE to JMB.