A computational model of topographic mapping and cell density in the macaque lateral geniculate nucleus.

摘要

In the primate visual system, nearby neurons in the brain represent nearby points in the visual field. These maps have been well studied in primary visual cortex (V1), but the lateral geniculate nucleus of the thalamus (LGN), which projects to V1, is far less understood. To date, efforts to study topography in the LGN have focused on one-dimensional models of magnification (the amount of neural substrate per degree of visual field), which cannot simply account for many important aspects of geniculate topography. Neither the presence of shear (angular distortion) nor differences in representation along radial segments within the visual field can be succinctly described with one-dimensional models. Efforts to study magnification are complicated by variable cell density in the LGN, which has not been well characterized. This thesis provides three contributions to the understanding of geniculate topography and magnification.;The first goal was to develop a two-dimensional model of topographic organization within the LGN. Extending the wedge-monopole model of V1 topography allows the development of a new model that jointly characterizes the six geniculate layers. Fitting this model to physiological data from macaque monkeys suggests that the geniculate displays an orderly increase in shear from innermost to outermost layers. The geniculate map structure is otherwise constant across layers.;The second goal was to develop a technique to address the variable cell density in LGN. The development of new methods to create density uniformizing maps (cartograms) allows a balance between highly-distorted cartograms and approximate cartograms with minimal shear. Applying these minimal shear transformations to the LGN enables direct comparison between LGN and V1 topography.;Finally, a manual count of neurons enabled investigation of the variability in LGN cell density across specimens. Counting a large number of cells demonstrates that previous estimates using different sample sizes are not comparable, and that commonly held views on geniculate cell distribution are likely to be incorrect.;These results demonstrate that characterization of geniculate topography is possible with a low-dimensional model, and that cell density (when correctly modeled) can be addressed in a principled way using conformal approximations to cartograms.