Functional Organization of Visual Pathways 1

Talk Session: Saturday, May 16, 2026, 2:30 – 4:15 pm, Talk Room 1
Moderator: Martin Hebart, Justus Liebig University Giessen

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

Talk 1, 2:30 pm, 24.11

Reorganization across cortical hierarchies and modality-specific thalamic plasticity in congenital sensory loss.

Monami Nishio¹ (), Xingyu Liu¹, Maria Zimmermann², Marcin Szwid³, Allyson Mackey¹, Michael Arcaro¹; ¹University of Pennsylvania, ²Johns Hopkins University, ³Jagiellonian University

Loss of sensory input leads to reorganization of cortical sensory systems. Here, we asked whether plasticity varies systematically with cortical distance from primary sensory areas, extends to higher-order thalamus, and follows similar principles across sensory modalities. Using structural and functional MRI, we characterized cortical and thalamic reorganization in adults with congenital blindness (N=15; controls N=19) or deafness (N=25; controls N=30). In blind individuals, lateral geniculate nucleus (LGN) volume and V1 surface area were significantly reduced and correlated across participants, suggesting coordinated thalamocortical structural reorganization. Beyond V1, changes in extrastriate surface area and morphology diminished with geodesic distance from V1. Visual cortex in blind individuals showed stronger resting-state correlations with the frontoparietal control network and weaker correlations with somatosensory cortex. This pattern was more pronounced in ventral than dorsal areas. Visual–control coupling was further enhanced during a decision-making task, suggesting recruitment of visual cortex for cognitive processing. Complementing these cortical changes, higher-order thalamic nuclei (dorsal pulvinar, central medial, mediodorsal) showed increased functional correlations with V1, indicating reorganization of thalamocortical circuits. Deaf individuals exhibited hierarchical plasticity in auditory cortex, with structural changes that diminished with distance from primary auditory cortex and increased cross-modal connectivity. However, unlike in blindness where deprived cortex coupled preferentially with frontal control regions, auditory cortex in deafness showed stronger coupling with visual cortex. Further, thalamic structures in deaf individuals showed no significant structural or functional alterations, indicating modality-specificity of thalamic plasticity from sensory loss. These findings reveal hierarchical principles governing cortical reorganization following early sensory loss and the modality-specific contribution of thalamic plasticity.

Talk 2, 2:45 pm, 24.12

Contributions of the primate superior colliculus to perceptual dominance during flash suppression

Divya Subramanian¹, Richard Krauzlis¹; ¹National Eye Institute, National Institutes of Health

We only consciously perceive a subset of our sensory inputs. How is this subset selected? We hypothesized that the primate superior colliculus (SC), important for selecting visual inputs to guide orienting, may play a role. To test this, we used binocular flash suppression to dissociate sensory inputs from the conscious percept. When an image presented to one eye is followed by an image added to the second eye, the second image becomes perceptually dominant even though both images are physically present. We asked whether visual activity in the macaque SC reflects the dominant image, the suppressed image, or the physically present mixed image. We presented oriented textures at two spatial frequencies that SC neurons distinguish at short latencies. We compared activity (n = 106 neurons from 2 monkeys) in the dichoptic, flash suppression condition to control conditions in which each image was unambiguously presented. We found that activity in the flash suppression condition matched the perceptually dominant image, regardless of whether the low or high spatial frequency image was dominant. However, SC activity only transiently reflected the dominant image. We trained linear classifiers to distinguish the two images based on population activity in the unambiguous control conditions and tested them either in the flash suppression condition or on held-out control trials. In the flash suppression condition, the classifier reliably labeled activity as belonging to the dominant image but only at short latencies (40 - 130ms from second image onset). For the held-out control trials, however, performance remained well above chance for the duration of the stimulus presentation (700 – 1000ms). Thus, SC activity transiently reflects the dominant image, suggesting that it reflects the perceptual switch rather than the sustained visual percept.

Talk 3, 3:00 pm, 24.13

Precise Functional Localization of the Foveolar Representations in Anesthetized Macaque Visual Cortex

Meizhen Qian^1,2, Meixuan Chen², Coco Laska^1,3, Anna Wang Roe^1,2; ¹Translational Neuroscience Laboratory Division, Center for Biomedical Imaging and Neuromodulation (C-BIN), Nathan S. Kline Institute for Psychiatric Research, ²Interdisciplinary Institute of Neuroscience and Technology, Zhejiang University School of Medicine, Hangzhou, China, ³Department of Medicine, Health, and Society, Vanderbilt University

Background: The primate foveola, which represents the central ~1° of the visual field, is critical for high-acuity vision and gaze-directed attention. In a cortical region previously called the foveal confluence, Qian (2025 Nature Neurosci) using ultrahigh field 7T MRI in awake fixating macaques discovered multiple foveolar loci encircling a large area termed the “foveolar core”, suggesting novel hierarchy for central vision. To study this foveolar area in anesthetized monkeys to determine exactly where the foveola is directed on a monitor, we introduce a systematic approach for foveolar localization. Methods: Functional EPI data were acquired at 0.8 mm³ isotropic resolution in a 7T MRI. Data acquisition was monocular. We used a progressive method of imaging with large to narrow stimuli to triangulate the center of gaze. Initially, wide (4.5°) vertical and horizontal bars containing drifting gratings were presented, at each of 3 different positions; the bar that induced the strongest response was chosen as the most central. Successively narrower bars (1.5° wide, then 0.5° wide) were presented at 3 positions within the central-most location. 0.2° spots were presented at the estimated foveal coordinate to further refine the central-most location. Results: Retinotopic shifts in activation corresponded well with stimulus position. The most central vertical bar stimulus activated the vertical meridian at the V1/V2 border; off-center bars produced parallel activations on either side of the border. When stimuli crossed into the ipsilateral hemifield, activation shifted to the other hemisphere as predicted. For horizontal meridian, the central-most stimulus produced the largest BOLD responses and activated dorsal and ventral V2/V3 borders. Foveolar spot stimuli exactly at the center produced bilateral foveolar activations. Significance: These results confirm the feasibility of precise foveolar localization in anesthetized macaques and provide a foundation for further studies of foveolar cortical organization using fMRI coupled with other recording and stimulation methodologies.

Talk 4, 3:15 pm, 24.14

Retinotopic Organization of Interhemispheric Structural Connectivity in Early Visual Cortex from Infancy to Adulthood

Seda Karakose-Akbiyik¹ (), Emily Kubota¹, Xiaoqian Yan², Sarah Tung¹, Christina Grace Tyagi¹, Danya Ortiz¹, Vaidehi Natu¹, Kalanit Grill-Spector¹; ¹Stanford University, ²Fudan University

Early visual cortex (EVC: V1, V2, V3) exhibits a retinotopic organization defined by polar angle and eccentricity, with each hemisphere representing the contralateral hemifield. Previous work suggests that the corpus callosum tracts linking the EVC in the two hemispheres are also retinotopically organized, with the densest connectivity for vertical meridian representations. Yet, in vivo evidence for these principles in humans remains limited, partly because early tractography methods lacked the precision needed to map these fine-grained connections. Recent advances in diffusion MRI (dMRI) now allow a more precise characterization of these connections. Here, we used dMRI to test whether the interhemispheric structural connectivity of V1–V3 is organized by retinotopy (i.e., polar angle and eccentricity). By examining both human adults (N=26) and infants at 0, 3, and 6 months (N=23, 23, and 21), we also tested how this organization emerges during infancy. We parcellated V1–V3 into polar-angle segments (V1-horizontal; V1–V2 lower/upper vertical meridians; V2–V3 dorsal/ventral horizontal meridians) and eccentricity bands (0–3°, 3–6°, 6–12°, 12–24°, 24–48°) using the Benson atlas, and identified the projection zones of each segment by mapping where streamlines from one hemisphere terminated in the other. In adults, contralateral projections of V1–V3 followed a clear retinotopic organization, consistent with the animal literature. Contralateral connections were denser along the vertical than the horizontal meridian, and lower/upper vertical-meridian segments primarily terminated in homotopic zones in the opposite hemisphere. Eccentricity also mattered: connections were denser between homotopic or adjacent eccentricities and sparser between distant ones. An adult-like organization was evident in infants, with both polar angle and eccentricity-based patterns already present for 0-month-olds. Together, these findings suggest that human V1-V3 callosal projections follow a retinotopic organization from birth. This organization likely plays a key role in coordinating signals across the two hemispheres to form a unified representation of visual space.

NIHRO1EY033835, NIHRO1EY023915

Talk 5, 3:30 pm, 24.15

Canonical spaces uncover shared and unique representational geometry of primate IT

Sander van Bree^1,2,3, Martin N. Hebart^1,2,3; ¹Justus Liebig University Giessen, ²Max Planck Institute for Human Cognitive and Brain Sciences, ³Center for Mind, Brain and Behavior

The inferotemporal cortex (IT) is central to high-level vision in primates, yet the extent to which its functional organization is conserved across species remains debated. In this study, we approach this question by systematically comparing macaque multi-unit activity and human fMRI responses to a large set of naturalistic images. Specifically, we used canonical correlation analysis (CCA) to construct three neural spaces, each of which provides complementary views on primate IT: macaque and human spaces obtained by aligning responses within species, and a universal cross-species space derived by combining all subjects. We demonstrate how these models can be used to uncover neural representations that are selective to species or generalize between primates, even with data obtained from different neuroimaging modalities. Analyzing these canonical spaces yields the following primary results. First, primate IT is high-dimensional and organized around shared axes, including but not limited to animacy and spiky-stubbiness. A non-negative, parts-based factorization of this universal space further uncovered a host of dimensions, ranging from grid-like patterns to biologically salient clusters related to faces and body silhouettes. Second, contrasting the species-specific spaces revealed a striking animacy asymmetry: the macaque space was driven largely by living categories, whereas the human space was better explained by non-living categories and human body parts. Furthermore, relative to the human space, the macaque space was more strongly captured by early visual representations, as evidenced by neural network activations and other predictors. Finally, the universal primate space held a shared representational structure for symbols, icons, and object ensembles, consistent with the hypothesis that human orthographic specialization builds on neural precursors already present in non-human primate IT. Together, this showcases how neural spaces derived from alignments within and across species provide a principled way to chart shared and species-specific representational geometry.

Supported by ERC StG COREDIM (101039712), LOEWE Start Professorship, and DFG: SFB/TRR 135 "Cardinal Mechanisms of Perception" (222641018- C11) & EXC 3066/1 "The Adaptive Mind" (533717223).

Talk 6, 3:45 pm, 24.16

Cut the tree: Image-grounded encoding models reveal distinct temporal profiles of naturalistic object and scene processing

Niklas Mueller¹ (), H. Steven Scholte¹, Iris Groen¹; ¹University of Amsterdam

It is well established that the brain achieves efficient visual information processing using specialized regions, such as scene-, face-, and object-specific regions which have been localized in humans using functional magnetic resonance imaging (fMRI). The temporal dynamics of the processing of different types of visual information have been understood less well. Some experiments use carefully designed stimulus presentations, such as flicker-stimuli, to study temporal information processing. However, these stimuli lack ecological validity to link temporal processing stages of visual perception to naturalistic, real-world visual information. Here, we use electroencephalography (EEG) recordings of human participants viewing large-field natural images and build brain-predictive image-grounded encoding models based on deep neural networks (DNNs) to study the temporal dynamics of object and scene processing. We separate object from scene information using manually created object bounding boxes and build separate encoding models for individual scene parts, such as different object classes (e.g., cars, tree, traffic signs) or scene elements (building, floor, sky). Our results show that object information is encoded delayed in time compared to early and rapid encoding of scene information. Moreover, with our approach, we can identify specific temporal encoding profiles for individual object and scene elements within a given naturalistic image. For example, we find that visual floor elements are maximally encoded around 150 ms, consistent with prior work showing the subsequent emergence of action affordance representations. Further, our encoding profiles show that while trees are commonly labeled as objects, they are processed more similarly to other scene elements than to objects. Overall, we establish a direct link between the pixel-wise information of ecologically valid stimuli and the temporal neural encoding profiles in EEG recordings. This opens new possibilities to study the neural processing of individual scene elements under more natural, real-world conditions.

Talk 7, 4:00 pm, 24.17

Spatiotemporal Continuity Shapes Information Flow in the Primate Ventral Stream

Alec G. Sheffield¹, Yuxuan Xue¹, Monika P. Jadi¹, Anirvan S. Nandy¹; ¹Yale University

Recurrent interactions between cortical areas are thought to be critical for coherent perception, yet how these long-range pathways depend on the spatiotemporal structure of natural vision remains unclear. We tested whether disrupting spatiotemporal continuity selectively impairs inter-areal communication in the ventral visual stream. We recorded simultaneously across cortical layers in areas V2 and V4 of awake marmosets (n=2) using laminar electrode arrays during free viewing. Animals watched intact movies or temporally shuffled versions in which consecutive 120-ms segments were randomly reordered, preserving local visual statistics while disrupting long-timescale spatiotemporal relationships. Spike-triggered current source density confirmed functional connectivity between recording sites. We quantified inter-areal communication using spike-field phase consistency, probabilistic graphical models (Multi-Timelag weighted Dynamic Bayesian Networks), and reduced-rank regression to characterize communication subspace geometry. Temporal shuffling exerted minimal effects on oculomotor behavior, firing rates, and population dimensionality, ruling out gross state differences as explanations for neural effects. However, inter-areal coordination was profoundly disrupted. Spike-field coupling decreased across theta frequencies in feedforward (V2→V4) and alpha/low-beta frequencies in feedback (V4→V2) directions. Critically, directed spiking dependencies revealed an asymmetric effect: feedback connections from deep-layer V4 to superficial V2 were selectively weakened, while feedforward dependencies remained intact. Subspace analyses showed that temporal discontinuity both rotated and destabilized the low-dimensional communication channels linking V2 and V4, reducing predictive accuracy and increasing geometric variability. In contrast, within-area laminar communication was largely preserved across conditions. These findings demonstrate a clear dissociation between local and long-range processing in the ventral stream. While local circuits can represent short-timescale visual information relatively independently, coherent coordination across hierarchical levels, particularly through feedback pathways, depends critically on the continuous spatiotemporal structure characteristic of natural vision.

Supported by NIH R01 EY034605, R00 EY025026, R21 MH126072, SFARI 875855 (MPJ); NARSAD YI, Ziegler Foundation, Yale Orthwein Funds; NIH R01 EY032555, R21 MH126072, SFARI 875855 (ASN); NSF GRFP (AGS); and NEI P30 EY026878 (Yale).