Spatial Vision: Neural mechanisms

Talk Session: Tuesday, May 20, 2025, 8:15 – 10:00 am, Talk Room 1

Schedule of Events | Search Abstracts | Symposia | Talk Sessions | Poster Sessions

Talk 1, 8:15 am

Evoked responses modulate perceptual sensitivity across visual space over time consistent with a cortical traveling wave.

Zachary Davis¹, Ashley Royston², Dylan Jensen¹, Emma Rudolph¹, Akshay Parchure¹; ¹University of Utah, ²University of Denver

Sensitivity for faint visual targets is not fixed but varies from moment to moment. One factor that has been proposed to influence visual sensitivity is the state of intrinsic fluctuations in subthreshold cortical activity that can travel as waves across the cortical surface (Davis, Muller, et al., Nature 2020). Sensory and motor events such as visual stimuli and saccades generate traveling waves of cortical activity (Muller et al., Nat. Communi. 2014; Zanos, et al., Neuron 2015) but it is unknown whether these waves also modulate sensitivity like intrinsically generated traveling waves. To test this, we asked healthy adult human subjects to report their detection of faint visual targets while triggering evoked responses (a visual stimulus or an eye movement) that putatively form traveling waves along a foveal-peripheral axis in the visual cortex. The targets appeared randomly at various eccentricities and delays with respect to the timing of the wave-triggering event and were titrated in size and contrast to achieve similar levels of performance. We tracked eye movements as subjects reported their detection of faint visual targets with a mouse button press. Detection performance followed a rhythmic time course with respect to trigger onsets consistent with previous reports (Fiebelkorn et al., Current Biology 2013; Hogendoorn, J. Cog. Neurosci. 2016). When we separated target performance based on target eccentricity, we found that the modulation time course shifted with eccentricity, following a progression consistent with the rhythmic fluctuation traveling as a wave. These results were recapitulated in a model that generated similar rhythmic performance modulations by activity propagation in the human visual cortex across the retinotopic map. This data is consistent with the view that spatiotemporal fluctuations in cortical activity modulate sensory processing and perceptual sensitivity and suggest that evoked activity fluctuations can impact sensitivity similar to intrinsically generated activity.

Talk 2, 8:30 am

Variability in hemodynamic response functions can masquerade as differences in retinotopic selectivity

D. Samuel Schwarzkopf^1,2 (), Ecem Altan¹, Catherine Morgan^3,4, Steven Dakin^1,5; ¹School of Optometry & Vision Science, University of Auckland, New Zealand, ²Experimental Psychology, University College London, United Kingdom, ³School of Psychology & Centre for Brain Research, University of Auckland, New Zealand, ⁴Centre for Advanced MRI, UniServices Ltd., Auckland, New Zealand, ⁵Institute of Ophthalmology, University College London, United Kingdom

Population receptive field (pRF) analysis has become the most popular method for retinotopic mapping with functional magnetic resonance imaging. It requires assumptions about the hemodynamic response function (HRF) to model a voxel’s response. Most pRF studies use canonical HRFs based on normative data or individual HRFs estimated via independent, event-related stimulation paradigms. However, the choice of HRF can influence pRF size estimates substantially (Lerma-Usabiaga et al., 2020, PLOS Comp. Biol. 16(6): e1007924). To understand how this might affect results in practice, here we concurrently fit a double-gamma HRF with five free parameters as part of the pRF model. Using simulated data, we first demonstrate that this algorithm accurately recovers different ground truth HRFs used for generating the data. Next, we reanalyzed several empirical pRF datasets collected using different stimulus paradigms (bar sweeps, combined wedge+ring), magnetic field strengths (1.5T, 3T, 7T), and pulse sequences. Concurrent HRF fitting affected pRF size estimates considerably. It typically improved the presumed validity of model fits by reducing the proportion of artifactually small pRFs. Importantly, the best-fitting HRF varied between datasets. It also differed substantially from all canonical HRFs and from HRFs measured independently, indicative of response non-linearities. Finally, HRF shape also differed substantially between early and higher visual cortex. All these differences could theoretically affect the interpretation of reported findings, such as claims about inhibitory modulation or contextual processing. To test this, we conducted further simulations. Even when the ground truth pRF had a circular Gaussian profile, analyzing these data with an inhibitory surround model (difference of Gaussians pRF) produced spurious estimates of surround inhibition. Such errors were greatly exacerbated when assuming a different HRF than the one used for data generation. Our findings therefore show pRF parameters estimated when assuming a fixed HRF should be treated with caution.

Supported by a Research Development Fund allocation from the Faculty of Medical & Health Sciences of the University of Auckland to DSS

Talk 3, 8:45 am

Decoding the orientation serial dependence effects from V1 neuronal responses using a transformer model

Xin Wang¹ (), Shi-Ming Tang¹, Cong Yu¹; ¹Peking University

When two lines are presented sequentially, the second line may be perceived as either slightly tilted towards the first one (attraction, influenced by Bayesian priors) or away from it (repulsion, driving by efficient coding), contingent on the experimental conditions. To understand the computational principles underlying these serial dependence effects, we employed two-photon calcium imaging to simultaneously record the responses of >1000 V1 superficial-layer neurons to sequentially presented Gabors embedded in white noise in two awake, fixating macaques. The parafoveal Gabor varied in 12 orientations and 4-5 contrasts (0.03-0.50), while the background noise varied in 5-6 RMS contrasts (0-0.29). The stimulus intervals were 1000-ms long, separated by 1500-ms inter-stimulus intervals. The average neuronal responses displayed maximum repulsion effects around 15 deg orientation differences. Consequently, downstream brain areas must readout V1 responses to generate attraction effects. We developed a transformer model that incorporated two self-attention mechanisms, each designed to identify the most relevant neurons and their effective connections for decoding the orientation of one stimulus line. Additionally, a cross-attention mechanism was included to capture interactions between the outputs of two self-attention mechanisms. The model was trained to reconstruct the actual repulsion effects and hypothesized attraction effects (repulsion effects flipped). A PCA analysis of the cross-attention maps corresponding to the outputs of the repulsion and attraction models revealed distinct distributions of neuronal responses within the top three PC dimensions. Moreover, within each distribution, cross-attention values of trial pairs with the largest effect sizes were clustered in the PCA space. These results suggest that the repulsive and attractive serial dependence effects in orientation perception are likely mediated by distinct and independent neural computations based on the same neuronal responses. Moreover, downstream brain areas can efficiently readout relevant V1 information to produce attraction effects, as indicated by the cross-attention values in the three-dimensional PCA space.

STI2030-Major Projects grant (2022ZD0204600)

Talk 4, 9:00 am

Characterizing population receptive fields in human visual cortex under wide-view stimulation

Pei-Yin Chen^1,2 (), Atsushi Wada^1,2; ¹Center for Information and Neural Networks, NICT, Osaka, Japan, ²Graduate School of Frontier Biosciences, Osaka University, Osaka, Japan

Population receptive field (pRF) analysis is a widely used method for investigating the retinotopic organization of human visual cortex. Yet, conventional fMRI visual presentation methods are typically limited to measuring pRF properties within 20 degrees of visual angle. Here, we utilized wide-view stimulation to explore the pRF properties of human visual cortex in both central and far-peripheral vision. The retinotopic stimuli consisted of rotating wedges and contracting concentric rings with checkerboard textures moving across the visual field. Observers viewed the stimuli through a wide-view binocular visual stimulation system (Wada et al., 2015, OHBM) in the MRI scanner, which allowed the visual presentation to cover around 90 degrees of the visual field. We applied a 2D Gaussian pRF model to the BOLD activations of voxels in areas V1 to V3, hV4, and V3A/B. Our results show a topographical representation of contralateral visual hemifield in all regions of interest, spanning both central and far-peripheral vision. When examining the relationship between the estimated pRF size and the eccentricity of the pRF center for both central and far-peripheral regions, we found that the relationship is better described by a logarithmic regression function rather than a linear regression function employed in previous studies which assessed only central visual regions. In addition, in areas V1 to V3, both the pRF center eccentricity and pRF size increased with the cortical distance from the posterior pole up to a certain level and then began to decrease. The range where the values increase can be captured by an exponential function, reflecting the cortical magnification in each area. On the other hand, the range where the values decrease may reflect potential limitations of the pRF estimation when the true pRF is located near the boundary of the visual presentation which can only be partially stimulated.

JSPS KAKENHI (Japan) 24K16880, 19K12745

Talk 5, 9:15 am

Visual response adaptation dynamics depend on luminance polarity and spatial frequency in primary visual cortex but not superior colliculus neurons

Yue Yu¹, João B. Bittar¹, Ziad M. Hafed¹; ¹University of Tuebingen

Stimulus-driven neuronal responses are often characterized by an initial spike burst followed by a gradual reduction towards lower steady-state activity. Such temporal adaptation is mathematically equivalent to high pass filtering, suggesting that different adaptation dynamics translate into different abilities to faithfully track time varying stimuli. Here, motivated by the idea that luminance polarity (dark versus bright contrasts) and spatial frequency may not be equally likely at all possible temporal frequencies in natural dynamic scenes, we asked whether primary visual cortex (V1) and superior colliculus (SC) neurons exhibit different adaptation time constants along these feature dimensions. For luminance polarity, we recorded from 408 V1 (two monkeys) and 238 SC (three monkeys) neurons. In each trial, a disc (0.51 deg radius and variable dark or bright Weber contrast) appeared within the neurons’ response fields. For spatial frequency, we recorded from 268 V1 and 127 SC neurons (two monkeys) while presenting a static Gabor grating (five spatial frequencies). SC neurons exhibited faster adaptation dynamics than V1 neurons. However, there was neither dependence on luminance polarity nor spatial frequency. In contrast, V1 neurons exhibited clear feature-dependence in their adaptation. For luminance polarity, adaptation was considerably slower for low contrast (10%) bright than dark stimuli. On the other hand, at 100% contrast, adaptation was considerably faster for brights than for darks. As for spatial frequency, V1 neurons always showed the fastest adaptation at intermediate (4 cycles/deg) spatial frequencies. We hypothesize that the faster SC adaptation allows visual-motor SC neurons to quickly recruit saccade-related bursts after the sensory responses. Instead, V1 neurons need to track scene dynamics. For example, clouds are low contrast bright stimuli that have slow temporal dynamics, not necessitating fast neural adaptation. Conversely, fixational eye movements enhance the temporal retinal image modulations of intermediate spatial frequencies, requiring faster V1 dynamics to represent them.

Talk 6, 9:30 am

Orientation selectivity in mouse superior colliculus modeled with center-surround receptive fields

Austin Kuo^1,2,3 (), Justin Gardner¹, Elisha Merriam^2,3; ¹Stanford University, ²National Institutes of Health, ³National Institute of Mental Health

Can a neural population be selective for properties of a stimulus none of its constituent neurons are selective for? Foundational single-unit physiology experiments show orientation-selective neurons emerging in primate V1 but not in subcortical structures such as LGN or superior colliculus (SC). Accordingly, a single-unit perspective suggests linear readout of stimulus orientation should be possible from V1, but not LGN or SC. We tested whether Ca2+ imaging results of orientation-selective populations from mouse SC necessarily implies SC neurons with V1-like, elongated receptive field (RF) structure, by simulating neural responses from a study (Liang et al., 2023, Nat Commun 14:4756) that measured mouse SC responses to visual stimuli varying in orientation (0, 15, …, 165°), spatial frequency (0.01-0.32 cycles/°), size (radii: 19, 29, 40°), and shape (circle, square, diamond). We simulated V1-like and center-surround RFs at single-unit and population scales, with RF sizes (~50-250 deg2) and spatial frequency selectivity (0.01-0.32 cycles/°) based on prior measurements. Reproducing empirical findings, neural populations simulated from either V1-like or center-surround RFs showed similar orientation preferences dependent on aperture location, shifting from radial to anti-radial orientation preferences as stimulus spatial frequency increased. At the single-unit scale, simulated center-surround, rather than V1-like RFs, matched empirical orientation preferences along stimulus apertures. Our simulations provide a unified explanation of edge-related orientation selectivity, suggesting measured orientation selectivity from SC populations need not imply V1-like RFs, but spatial frequency selectivity in single units can instead confer orientation selectivity to the population. This result, however, does not preclude existence of V1-like RFs in prior empirical data. Rather, our simulations provide a tool for experimenters to determine RF properties sufficient for producing measured activity. Broadly, our results demonstrate that neural populations can exhibit emergent selectivity absent in their constituent neurons, which in principle could be read out by downstream perceptual and motor systems.

ZIAMH002966, Wu Tsai Neurosciences Institute