Abstract
When a stimulus is viewed twice, neural responses in the visual cortex tend to be lower to the second image, known as repetition suppression (RS). Previous work showed that a model of divisive normalization (DN model) can predict the level and time course of short-term RS in neural responses in the visual cortex to consecutively presented contrast patterns (Zhou et al., 2019, PLoS Comput Biol; Groen et al., 2021, biorXiv). While the DN model accurately predicts responses to simple contrast patterns, it does not account for spatial properties of the stimulus, and therefore cannot make different predictions as a function of stimulus category. For complex natural images, the stimulus category may affect the degree of RS, particularly in higher visual areas, where image class strongly influences neural responses, also known as category-selectivity. Here we show that a modified version of the DN model that incorporates category-selectivity is able to predict RS in both early and late visual areas. Time-varying broadband time series data (50-200Hz) was analysed from 102 visually responsive electrodes across 4 patients with epilepsy undergoing electrocorticographic recordings. Subjects were presented with two identical, naturalistic images belonging to one of six categories (bodies, buildings, faces, objects, scenes, scrambled) with a varying inter-stimulus interval (17-533ms). Results show, first, higher areas (beyond V1-V3) exhibit a stronger level of RS relative to early areas (V1-V3). Second, within higher areas, RS was stronger for preferred than non-preferred image categories, with responses to preferred stimuli exhibiting a slower time-to-recovery. Lastly, incorporating category-selectivity into the model explained ~20% more variance in the cross-validated time series than the normalization model without category-selectivity. Together, these results reveal differences in temporal response dynamics across the visual hierarchy during processing of natural images and offer a computational approach to predict adaptation in the human brain at the millisecond scale.