Complex Pattern Selectivity in Macaque Primary Visual Cortex Revealed by Large-Scale Two-Photon Imaging

Abstract

Visual objects contain rich local high-order patterns such as curvature, corners, and junctions. In the standard hierarchical model of visual object recognition, V1 neurons were commonly assumed to code local orientation components of those high-order patterns. Here, by using two-photon imaging in awake macaques and systematically characterizing V1 neuronal responses to an extensive set of stimuli, we found a large percentage of neurons in the V1 superficial layer responded more strongly to complex patterns, such as corners, junctions, and curvature, than to their oriented line or edge components. Our results suggest that those individual V1 neurons could play the role in detecting local high-order visual patterns in the early stage of object recognition hierarchy.

Convolutional neural network models of V1 responses to complex patterns

“…Cell classification The recorded neurons in the neural data had mixed tuning properties (Tang et al, 2018): some acted more like complex pattern detectors, some acted more like simple oriented edge detectors, and some had weak responses to all the presented stimuli. To allow cleaner and more interpretable model comparisons, we evaluated model performance for different types of neurons separately (Section 5).…”

“…To make such per-neuron-type comparison possible, a classification of neurons is required. Here we use the neuron classification scheme in Tang et al (2018). First, neurons whose maximum mean responses were not above 0.5 (max r n t ≤ 0.5) were discarded as their responses were too weak and might be unreliable; then, among all the remaining neurons that passed the reliability test, neurons whose maximum mean responses over nonOT stimuli were more than twice of those over OT stimuli ( max r n t 1 max r n t 2 > 2, where t 1 and t 2 go over all nonOT and OT stimuli respectively) were classified as HO (higher-order) neurons and the others were classified (conservatively) as OT neurons; finally, all the HO and OT neurons were further classified into subtypes, such as curvature neurons and corner neurons, based on ratio tests similar to the one above-for example, an HO neuron was additionally considered as a curvature neuron if its maximum response over curvature stimuli was more than twice of that over noncurvature stimuli.…”

“…Recording technique While most other studies use spiking data collected using multi-electrode array (MEA) technologies, we use calcium imaging data (Li et al, 2017;Tang et al, 2018). Although MEA-based spiking data are in theory more accurate, calcium imaging techniques can record many more neurons and do not suffer from spike sorting errors.…”

“…Second, more sophisticated analysis techniques have revealed richer structures in V1 neurons than those dictated by standard models Carandini et al, 2005). As an additional yet novel demonstration of this gap, using large-scale calcium imaging techniques, we (Li et al, 2017;Tang et al, 2018) have recently discovered that a large percentage of neurons in the superficial layers of V1 of awake macaque monkeys respond strongly to highly specific complex features; this finding suggests that some V1 neurons act as complex pattern detectors rather than Gabor-based edge detectors as dictated by classical studies (Jones and Palmer, 1987a;Dayan and Abbott, 2001).…”

Stimulus Vignetting and Orientation Selectivity in Human Visual Cortex

“…Stimulus vignetting is a general issue of concern in visual neuroscience (Carter and Henning, 1971;Cichy et al, 2015;Tang et al, 2018). According to the theoretical model that we have applied here, whenever a neuron's receptive field overlaps a stimulus edge or a change in contrast, stimulus vignetting will spread the Fourier power and affect the neuron's response.…”