Prenatal development of ocular dominance and orientation maps in a self-organizing model of V1
Active from 2004 - 2005

The primary visual cortex (V1) is organized into maps of ocular dominance (OD) and orientation (OR). It is known that these maps are altered by visual experience. However, the structures already exist before the newborn cat, ferret or monkey opens its eyes. The mechanisms of this previsual map development have been a controversial issue. Possible influences include molecular signals and spontaneous neural activity, but their respective contributions remain unclear.

With the help of the LISSOM model of the primary visual cortex the effect of different patterns of spontaneous activity has been investigated. The results suggest that previsual spontaneous activity alone is sufficient for realistic OR and OD maps to develop. Maps of only one feature, i.e. OR or OD alone, develop robustly with a wide range of activity patterns. However, joint OR/OD maps depend crucially on how correlated the patterns are between the eyes, even over brief initial periods. Therefore the simultaneous measurement of joint maps and map interactions rather than of maps of a single feature may reveal more information about the biological processes of previsual map development.

The figures that are referred to in the following are available in a PDF document.

OR and OD maps and their interactions

Previsual maps in animals have four main properties: (i) smooth, binocularly matching OR maps, (ii) an OD map with significant left or right eye preferences and smooth transitions between them, (iii) orthogonal intersections of OR and OD maps, and (iv) OR pinwheel centers located within OD columns.

Various simulations tested how well the maps generated by different neural activity patterns match the properties of animal maps. In the extreme case of very strong input correlation between the eyes, matching OR maps develop but no ocular dominance map (Fig. 3a). Conversely, in the extreme case of very weak correlation between the input patterns, strong OD maps develop, but the OR maps are different in each eye (Fig. 3b). Thus spatial correlation of the input is required for property matching OR maps, yet some degree of uncorrelation is simultaneously needed for the development of ocular dominance.

Maps that match properties (i)-(iv) developed with a combination of correlated and uncorrelated activity patterns. After the first 20% of the development, the ratio of correlation was decreased from 1:1 to 1:2 (Fig. 4).

The effect of noise on selectivity in OD map development

Prior LISSOM experiments demonstrated that uncorrelated patterns without noise lead to very strong selectivity in OD maps, similar to maps of adult strabismic animals. Those results suggest that uncorrelated spontaneous activity patterns could result in the earliest maps being strabismic even in normal animals. However, we found that including an overall pattern of noise remedies the extreme selectivity. This equalizing effect even persists long after the noise has disappeared. In the maps presented in Figure 5, noise was only present during the first tenth of the development.

Prenatal Development of Ocular Dominance and Orientation Maps in a Self-Organizing Model of V1 Stefanie Jegelka, James A. Bednar, and Risto Miikkulainen Neurocomputing, 69:1291--1296, 2006. 2006