We ended Part 1 with a pledge to share the highlights of Chapter 21 of Kandel and colleagues’ Principles of Neural Science (The Constructive Nature of Visual Processing) in part two, so here we go, in their words:
- Vision is a constructive process fundamentally different from the mere recording of visual input as in a camera. Rather, the brain uses visual input to infer information about the world around it, including information about objects, such as their sizes, shapes, distances, and identities and how rapidly they are moving.
- The tuning of neural circuits for visual features such as contrast, orientation, and motion often matches the distribution of the feature in the natural environment. This suggests an evolutionary, ethologically driven origin for the neural circuitry.
- Visual circuitry, and thus vision, are modulated by individual visual experience.
- Vision makes extensive use of parallel processing. The higher visual centers form two distinct pathways. The dorsal pathway, located in parietal cortex, is involved in motion perception, attention, and visually guided action. The ventral pathway, located in temporal cortex, processes form and objects. Further subdivisions of the ventral pathway are specialized, for example, for recognizing faces. These pathways, although distinct, communicate with each other; this is likely important for the perception of objects as coherent wholes.
- Parallel processing starts at the retina. Distinct retinal circuits analyze each point of the visual input for different local features including local contrasts of achromatic bright versus dark, red versus green, and blue versus yellow. The information is sent out through distinct classes of retinal ganglion cells (magnocellular, parvocellular, and koniocellular, respectively, for the three features noted) whose axons form the optic nerves.
- The optic nerves from the two eyes regroup at the optic chiasm such that all the fibers from the left visual hemifield project to the right hemisphere of the brain, and vice versa. However, the parallel retinal channels remain anatomically segregated by eye and by visual feature, past a thalamic relay station, the lateral geniculate nucleus (LGN), up to primary visual cortex (V1).
- The different channels enter V1 at different layers, although primarily they enter at the major input layers 4 and 6. The visual input is recombined to extract new sets of features. These include tuning for orientation, motion, and object depth (obtained by combining left and right eye inputs).
- V1 neurons sharing basic properties such as spatial location or orientation preference form columns extending vertically from the pia to the white matter.
- V1 neurons also form systematic horizontal maps of their response properties over cortex. The tuning for location forms a smooth “visuotopic” map of visual space which changes gradually with distance, and is most finely resolved at the fovea, growing progressively coarser toward the periphery. Superimposed on the spatial map are locally smooth maps of orientation preference and left versus right eye preference, with interspersed columns that preferentially process color. These visual response features cycle over relatively short cortical distances, in effect completing one full cycle over each partial shift of the spatial map. Thus, V1 circuits effectively analyze each visual location, in parallel, for the full set of V1 visual features.
- Neural processing in V1 reflects its architecture, with local vertical processing along columns and lateral processing across columns. In addition, there is long-range processing that spans multiple columns.
- The output of V1 feeds into progressively higher visual areas comprising more than 30 centers distributed along the dorsal and ventral pathways. The connectivity is reciprocal, with higher loci sending dense feedback targeting lower areas including the LGN.
- A useful measure of visual processing is provided by changes in neuronal “receptive fields” along the visual pathway. The receptive field is the region of visual space from which the neuron receives input; it is further characterized by the neuron’s optimal visual stimulus. Receptive fields grow larger and more complex at successive stages along the visual pathway. Their optimal stimuli also increase in complexity from simple pixel-like dots for photoreceptors, to oriented lines for V1, to face in higher face-selective centers of the ventral pathway.
- Looking forward, one of the most important unresolved questions is the interaction between feedforward visual processing through progressively “higher” neural computations and feedback mediated via the dense plexus of connections from higher to lower levels. Understanding this interaction may be the key to understanding how the brain effortlessly forms complex visual shapes.
In Part 3 we’ll begin to answer this and other key questions by taking a look at what Kandel and colleagues identify as low-level visual processing, originating at the retinal level. If nothing else, reading Principles provides renewed appreciation for the complexities of normal visual development, and the vulnerability of these processes to brain injury.