Comprehensive Study Notes on Visual System Physiology, Retinal Microcircuitry, and Cortical Processing Pathways

Visual Resolution, Spatial Sampling, and Foveal Cone Architecture

The fundamental limits of human spatial vision are established by the density and spatial arrangement of photoreceptors in the central fovea. Visual discrimination in human observers reaches a fine resolution threshold of approximately $1\text{ min of arc}$. This behavioral threshold directly corresponds to the physical center-to-center spacing of approximately $3\,\mu\text{m}$ between adjacent cone photoreceptors in the central foveola mosaic (Wässle & Boycott, 1991).

The spatial sampling of visual stimuli across the foveal mosaic can be modeled by analyzing the light intensity profile projected onto the cone array. When a single thin line of light is projected onto the retina, it produces a narrow gaussian-like illuminance profile. When two thin lines separated by $1\text{ min of arc}$ are presented, the combined light profile displays two distinct peaks separated by a central dip. This modulation in light intensity across adjacent cones provides the physiological signal required to discriminate two separate visual lines.

Spatial resolution limits for grating stimuli depend on the spatial wavelength $\lambda$ of the grating relative to the cone center-to-center distance $a$. When $\lambda = 2a$, spatial sampling satisfies the Nyquist criterion along orthogonal rows of cones, allowing correct spatial frequency reconstruction without aliasing. When $\lambda = a\sqrt{3}$, the grating spatial frequency aligns with the diagonal hexagonal packing geometry of the cone array. When $\lambda = a$, the spatial period equals the inter-cone spacing, causing severe aliasing and loss of spatial pattern resolution.

Functional Segregation in Early Pathways: Parvocellular, Magnocellular, and Koniocellular Systems

Visual processing beyond the photoreceptor layer is organized into parallel, functionally distinct streams. Experimental lesion studies in the non-human primate lateral geniculate nucleus (LGN) demonstrate clear anatomical and functional segregation between the parvocellular and magnocellular pathways (Merigan, 1989; Merigan & Maunsell, 1990; Schiller et al., 1990a,b).

Selective lesions placed within the parvocellular layers of the monkey LGN result in a severe, highly specific loss of wavelength-based color discrimination and a marked reduction in contrast sensitivity for high spatial frequencies. However, these parvocellular-lesioned animals maintain completely normal detection thresholds for high-frequency temporal flicker. Conversely, selective lesions within the magnocellular layers preserve wavelength-based color discrimination and high spatial frequency contrast sensitivity, while producing a profound impairment in high-frequency temporal flicker resolution and reduced contrast sensitivity at low spatial frequencies. These lesion findings establish that the parvocellular system mediates high-acuity form perception and color vision, whereas the magnocellular system processes motion, high temporal frequencies, and low spatial frequency contrast.

Single-unit electrophysiological recordings reveal fundamental physiological differences between parvocellular and magnocellular neurons in the LGN. Parvocellular neurons display color-opponent receptive field organization, sustained temporal firing patterns, tuning to low temporal frequencies, slower axonal transmission velocities, linear spatial summation, high spatial frequency sensitivity, relatively weak contrast sensitivity, and direct axonal projections to layer $4C\beta$ of primary visual cortex (V1). Magnocellular neurons exhibit weak or absent color opponency, transient temporal responsiveness, tuning to high temporal frequencies, rapid axonal transmission velocities, fairly linear spatial summation, low spatial frequency sensitivity, high contrast sensitivity that saturates at low contrast levels, and direct axonal projections to layer $4C\alpha$ of V1.

Isoluminant gratings, which consist of alternating colored bars (such as red and green) matched precisely in luminance, are used to functionally isolate the parvocellular system (Livingstone & Hubel, 1987, 1988). Because the spectral sensitivity of magnocellular neurons closely matches the photopic luminance function ($V_\lambda$), a green bar and a red bar of identical photopic luminance activate magnocellular cells to the exact same degree. Consequently, the border between red and green bars becomes invisible to magnocellular neurons, effective silencing the magnocellular pathway. Visual processing of isoluminant stimuli is mediated almost exclusively by the isolated parvocellular system, leading to perceptual phenomena such as impaired motion perception and altered depth perception.

Retinal Microcircuitry and Parallel Bipolar-Ganglion Networks

The architectural organization of retinal circuits changes systematically with retinal eccentricity. In the central retina and parafovea within $4\,\text{mm}$ of the foveal center, high spatial resolution is maintained through an un-converged midget pathway. In this central region, $1\text{ cone}$ connects to $2\text{ midget bipolar cells}$—one invaginating midget bipolar cell (imb1, ON-type) and one flat midget bipolar cell (fmb1, OFF-type)—which in turn project to $2\text{ midget ganglion cells}$ (one ON-center midget ganglion cell and one OFF-center midget ganglion cell).

In accordance with laminar organization rules in the inner plexiform layer (IPL), OFF-center midget bipolar and ganglion cell dendrites arborize in sublamina a (outer IPL, near the inner nuclear layer), whereas ON-center midget bipolar and ganglion cell dendrites arborize in sublamina b (inner IPL, near the ganglion cell layer) (Nelson et al., 1978; Dacey et al., 2000). Anatomical studies demonstrate that L-cone and M-cone midget bipolar terminal axons differ in their synaptic ribbon numbers, indicating potential differences in synaptic gain between red and green cone pathways (Calkins et al., 1994).

At greater retinal eccentricities, spatial convergence increases dramatically. At $4\,\text{mm}$ from the fovea, multiple midget bipolar cells (imb1, imb2, imb3) converge onto individual midget ganglion cells.

At $12\,\text{mm}$ from the fovea, dendritic fields expand further, with extensive synaptic convergence from multiple midget bipolar terminals and dense amacrine cell networks providing inhibitory input onto ganglion cell dendrites.

Three distinct bipolar cell classes drive the parallel visual pathways into the brain. Midget bipolar cells receive input from single L- or M-cones to establish red-green color opponency and high spatial acuity, projecting exclusively to midget ganglion cells that form the origin of the Parvocellular LGN pathway. S-cone bipolar cells synapse selectively onto sparse blue (S) cones to encode blue-yellow opponency, projecting to small bistratified ganglion cells that form the Koniocellular LGN pathway. Diffuse bipolar cells collect inputs from many cones and rods, displaying no color opponency and high light sensitivity due to spatial pooling; they project to parasol ganglion cells, establishing the Magnocellular LGN pathway.

Antagonistic surround receptive fields in bipolar and ganglion cells are generated by lateral feedback connections mediated by horizontal cells (HI, HII, and HIII types). HI horizontal cells connect predominantly to L- and M-cones, providing local lateral inhibition that structures center-surround organization. HII horizontal cells contact S-cones, contributing to blue-yellow chromatic opponency networks.

Color-opponent ganglion cells in the primate retina display discrete spatial-chromatic receptive field configurations (Gouras, 1968). These include Red ON / Green OFF center-surround cells, Red OFF / Green ON center-surround cells, Green ON / Red OFF center-surround cells, Green OFF / Red ON center-surround cells, and co-extensive Blue ON / Yellow OFF cells.

Mammalian retinas contain extensive morphological diversity across all five major cell classes: photoreceptors (rods and cones), horizontal cells, bipolar cells, amacrine cells, and ganglion cells. Amacrine cells display the greatest structural and functional diversity, featuring dozens of morphological subtypes that mediate complex intra-retinal processing.

Rod Pathways and Low-Light Scotopic Processing

Scotopic vision relies on extreme anatomical convergence to detect low light levels. Multiple rod photoreceptors pool their signals onto specialized rod bipolar cells ($\text{RB}$), with $15\text{ to }30\text{ rods}$ contacting a single rod bipolar cell in the outer plexiform layer (OPL).

Quantitative convergence ratios highlight the massive spatial pooling present in scotopic circuits. In highly convergent peripheral pathways, $75{,}000\text{ rods}$ converge onto $5{,}000\text{ rod bipolars}$, which synapse onto $250\text{ AII amacrine cells}$ to drive a single alpha ganglion cell ($\text{alpha GC}$). In less convergent pathways, $21{,}000\text{ rods}$ converge onto $140\text{ rod bipolars}$ and $7\text{ AII amacrine cells}$ to drive a single beta ganglion cell ($\text{beta GC}$).

Rod bipolar cells do not form direct synaptic connections with retinal ganglion cells. Instead, rod bipolar axon terminals form excitatory glutamatergic chemical synapses onto AII amacrine cells and A17 amacrine cells in sublamina b of the IPL. The AII amacrine cell serves as the central hub of the scotopic network: it forms depolarizing electrical gap junctions with the axon terminals of ON-center cone bipolar cells (cb5), which subsequently excite ON-center ganglion cells. Simultaneously, the AII amacrine cell makes glycinergic inhibitory chemical synapses onto the axon terminals of OFF-center cone bipolar cells and directly onto OFF-center ganglion cells.

Parallel retinal circuits process light increments (brightness) and light decrements (darkness). Brightness processing is initiated by depolarizing bipolar cells that depolarize in response to light onset, exciting ON-center ganglion cells. Darkness processing is driven by hyperpolarizing bipolar cells that depolarize to light offset, exciting OFF-center ganglion cells. Dopaminergic amacrine cells (A18) and interplexiform cells (IPC) provide modulatory feedback across these scotopic networks.

Retinal Feature Detection and Non-Linear Visual Computation

Electrophysiological discoveries by Stephen Kuffler, Ragnar Granit (1967 Nobel Laureate), and H. K. Hartline demonstrated that extensive visual processing and filtering occur directly within the retina. Retinal ganglion cells act as specialized feature detectors that divide visual input into parallel processing channels (Lettvin et al., 1959).

In the frog retina, specific ganglion cell classes are tuned to distinct ecological features. One class, termed "bug detectors", responds selectively to small, dark, airborne spots moving across the visual field while ignoring stationary or diffuse light patterns. Another class responds exclusively to large, moving shadows, signaling the overhead approach of predators.

Directionally selective ganglion cells (DSGCs) discharge action potentials when a visual stimulus moves in a preferred direction, but are inhibited when the stimulus moves in the opposite (null) direction (Barlow, Hill, & Levick, 1964). Preferred direction selectivity is independent of object contrast (occurring for both bright and dark spots) and stimulus form (occurring for single spots or moving gratings). Directionally selective cells are also tuned to specific movement velocities.

ON-OFF DSGCs possess bi-stratified dendritic arbors in sublaminae S2 and S4 of the IPL, receiving excitatory inputs from ON and OFF bipolar cells and asymmetric inhibitory inputs from starburst amacrine cells (SACs). Movement in the null direction evokes localized starburst inhibition that quenches excitation before it reaches the ganglion cell soma.

Y-type retinal ganglion cells exhibit non-linear spatial integration that makes them sensitive to texture motion (Hochstein & Shapley, 1976). When a fine spatial grating shifts in either direction over the receptive field, the Y-cell generates a transient burst of action potentials, even though total mean illumination over the receptive field remains unchanged. This sensitivity is created by underlying bipolar cells that possess biphasic temporal dynamics and non-linear synaptic rectification, ensuring that depolarizing phases drive transient excitation during grating shifts.

During steady visual gaze, fixational ocular drift and microsaccades shift visual images over foveal photoreceptors across displacements of approximately $5\text{ arcmin}$, refreshing retinal input and preventing neural adaptation.

Object Motion Sensitive (OMS) ganglion cells distinguish motion of an isolated object from self-induced motion of the entire background visual scene. When motion occurs exclusively in the receptive field center, bipolar cells deliver a sequence of excitatory inputs to the OMS ganglion cell, driving vigorous spiking. When the background moves synchronously with the center (global motion), wide-field polyaxonal amacrine cells are activated across the surround. These polyaxonal amacrine cells send inhibitory signals that act pre-synaptically on bipolar cell terminals, suppressing excitatory output and silencing the OMS ganglion cell. When center motion differs from background motion (differential motion), pre-synaptic inhibition fails to align temporally with center excitation, allowing excitatory signals to reach the OMS ganglion cell and drive action potentials.

Retinal circuits also contain specialized detectors for looming or approaching motion (Münch et al., 2009). Approaching dark objects evoke strong spike discharges in specialized ganglion cells, whereas lateral object movement elicits weak firing. Approaching motion sensitivity is generated by an excitatory input from OFF bipolar cells combined with non-linearly rectified inhibition from ON bipolar-driven amacrine cells coupled via gap junctions.

Structural and Functional Architecture of Primary Visual Cortex (V1)

Primary visual cortex—also designated as striate cortex, visual area 1 (V1), and Brodmann area 17—is located within the occipital lobe of the cerebral cortex. It is named striate cortex due to the presence of a prominent stripe of myelinated geniculate axons located in cortical layer $4B$, known as the Line of Gennari. V1 contains six neocortical layers, with layer 1 situated most superficially and layer 6 situated deepest.

Visual signals pass from retinal ganglion cells through the optic nerve and optic tract to synapse in the LGN. LGN relay neurons send optic radiations to V1. Inputs from parvocellular LGN layers synapse in layer $4C\beta$ of V1, whereas inputs from magnocellular LGN layers synapse in layer $4C\alpha$. Layer 6 neurons send reciprocal feedback projections back to the LGN and pulvinar nucleus, whereas superficial layers 2 and 3 send efferent projections to extrastriate cortical areas.

David Hubel and Torsten Wiesel (1962, 1968, 1977) categorized V1 neurons into distinct functional classes based on receptive field properties. Simple cells possess receptive fields partitioned into separate, spatially antagonistic excitatory ($+$, ON) and inhibitory ($-$ or $action$, OFF) subregions. Simple cells respond maximally to stationary or moving bars or edges oriented at a specific spatial angle. Presenting a bar at non-preferred orientations drastically reduces or eliminates the neural response. Hubel and Wiesel proposed that simple cell receptive fields are constructed by the linear convergence of multiple LGN neurons whose concentric ON- or OFF-center receptive fields are arranged along a straight line in visual space.

Complex cells display orientation and direction selectivity without distinct spatial ON and OFF subregions. A complex cell fires continuously as an appropriately oriented light or dark bar moves across its receptive field. Many complex cells exhibit directional selectivity, firing vigorously to movement in one preferred direction and remaining silent to movement in the opposite direction.

End-stopped cells (formerly termed hypercomplex cells) show length tuning. An end-stopped cell fires maximally when an oriented bar matches the precise length of its receptive field center. Extending the bar length beyond this optimal threshold engages inhibitory end-zones, suppressing neural firing.

Striate cortex is organized into functional modules termed hypercolumns, which integrate three structural systems across layers 1 through 6. Ocular dominance columns consist of vertical slabs of cortex that respond preferentially to input from either the right eye or the left eye. Orientation columns consist of vertical slabs containing neurons tuned to identical stimulus orientations. Penetrating an electrode perpendicular to the cortical surface encounters cells with identical orientation tuning. Penetrating an electrode obliquely across layers 2 and 3 reveals a systematic shift in orientation preference. Higher-resolution imaging reveals that orientation preferences rotate continuously around central singular points in a pinwheel configuration (Blasdel, 1992). Neurons in layer 4 lack orientation selectivity, displaying concentric receptive fields similar to LGN cells.

Cytochrome oxidase histochemistry reveals cylindrical structures termed blobs distributed across layers 2 and 3 of V1. Blob neurons lack orientation selectivity and contain double color-opponent receptive fields (for example, Red ON / Green OFF in the center, and Red OFF / Green ON in the surround). Optimal activation of a double color-opponent blob cell requires high chromatic contrast, such as a red center spot surrounded by a green annulus. Interblob regions contain orientation-selective, achromatic simple and complex cells.

Ocular dominance columns refine during early post-natal development through activity-dependent Hebbian plasticity mechanism. In dark-reared or infant primates, LGN axon terminals from both eyes overlap extensively in layer 4. Over a critical period spanning 2 weeks to 13 weeks after birth, synchronous driven activity segregates left-eye and right-eye LGN terminals into distinct alternating stripes. Monocular visual deprivation during this critical period causes axon terminals driving the non-deprived eye to expand, while terminals driving the deprived eye retract, leaving the deprived eye with reduced cortical representation.

High-Level Cortical Processing Streams and Object Recognition

Visual processing beyond V1 separates into two cortical processing streams (Mishkin et al., 1983; Maunsell & Newsome, 1987; Merigan & Maunsell, 1993; Milner & Goodale, 1995). The ventral processing stream (the temporal pathway, or "what" system) originates from parvocellular inputs projecting via V1 layer $4C\beta$, cytochrome oxidase blobs, and interblobs to thin and pale dark stripes in visual area 2 (V2), then to visual area 4 (V4) and inferotemporal cortex (IT). The ventral stream processes color, high spatial detail, complex visual forms, and object identity.

The dorsal processing stream (the parietal pathway, or "where" system) originates from magnocellular inputs projecting via V1 layer $4C\alpha$ and layer $4B$ to thick dark stripes in V2, then to middle temporal area MT (V5) and parietal area PG. The dorsal stream processes motion perception, spatial localization, and visually guided motor control. Reciprocal feedback connections link higher extrastriate areas back to V1 and lower visual centers, providing top-down attentional modulation and receptive field gating (Shapley, 2004).

The dorsal stream can be further subdivided into two functional pathways (Rizzolatti & Matelli, 2003). The dorsal-dorsal stream processes action execution; clinical damage to this pathway causes optic ataxia, characterized by an inability to accurately reach for or point to visually presented objects. The ventral-dorsal stream (which includes area MT) processes spatial perception and motor action planning.

Inferotemporal cortex (IT in macaques, equivalent to the lateral occipital complex / LOC in humans) contains neurons with large receptive fields that integrate spatial information across extensive visual angles (Gross, 1973; Fuster & Jervey, 1981; Rolls & Tovée, 1995). IT neurons display position and size invariance, responding to complex geometric shapes such as circles, squares, ovals, and star patterns (Tanaka et al., 1991; Gallant et al., 1993). Functional MRI studies confirm that human LOC responds robustly to intact 3D objects, but shows minimal activation to scrambled objects or disconnected shape fragments (Grill-Spector et al., 2001).

Single-unit studies using composite shape stimuli ("batons") demonstrate that IT neurons are tuned to specific combinations of shape sub-components. When testing an array of 64 baton stimuli systematically varying in top and bottom parts, an IT neuron fires maximally to the precise pairing of its preferred top feature and preferred bottom feature.

Single-cell recordings within fMRI-identified face patches in primate IT cortex reveal specialized face processing modules (Tsao et al.). Across 320 recorded neurons, cells display extreme selectivity for face images compared to bodies, fruits, gadgets, hands, or scrambled images. Minimal feature dissection reveals that the critical visual arrangement driving face-selective IT neurons consists of two dark spots and one horizontal dark bar positioned within a dark circular boundary (Kobatake & Tanaka, 1994). Faces are processed configurally as holistic visual patterns, whereas non-face objects (such as houses) are processed by parsing individual structural parts.

Lesions within specific extrastriate regions result in visual agnosias. Object agnosia impairs the recognition of real physical objects; agnosia for drawings impairs the recognition of line drawings; prosopagnosia causes an inability to recognize familiar faces; simultagnosia prevents perceiving more than one object at a time; color agnosia impairs associating correct colors with objects; color anomia prevents naming colors; cerebral achromatopsia destroys hue discrimination; visual spatial agnosia impairs stereoscopic depth perception and topographical orientation; and cerebral akinetopsia causes motion blindness.

Outputs from IT cortex project to medial temporal lobe (MTL) structures, including the hippocampus and amygdala. Single-neuron recordings in human MTL patients reveal highly selective, perceptually invariant neurons termed concept cells. Individual MTL neurons discharge selectively to multiple diverse visual images of a specific person or landmark (such as Diego Maradona, Brad Pitt, Luke Skywalker, or the Sydney Opera House), as well as to their printed or spoken names. Concept cells form sparse cell assemblies with partially overlapping neural representations, providing the circuit mechanism for encoding semantic associations and episodic memories.