Principal components analysis of cells in cat visual cortex

Abstract

Cortical cells differ along many dimensions, and statistical techniques are needed to study the relationships among them. We have used principal components analysis to study variation within a population of cells in the visual cortex of the cat. This procedure transforms the data which may then be projected onto a small number of independent axes. The axes are hierarchically ordered in terms of the sample variance they account for so that subsequent components account for decreasing proportions of the total sample variance. Four principal components (PCs) were identified by the analysis. The contributing variables on these components were respectively: (1) cutoff velocity, peak response, receptive field area and spontaneous activity; (2) tuning width, orientation selectivity, occular dominance ratio and obliquity; (3) directionality ratio, monocularity and vvariability; and (4) width to length ratio, variability and sideband ratio. We hypothesize that each of these components identifies a physiologically important feature of cortical organization. PC 1 and PC 4 describe those relationships among variables that have been used previously by other workers to classify cells in the visual cortex of the cat. Specifically, PC 1 is highly correlated with those variables associated with type of afferent input and PC 4 is highly correlated with those variables that differentiate simple cells from complex cells. PC 2 and PC 3, however, are highly correlated with variables that have not been used in classiying cells. Our results demonstrate that: (1) each of two major classification schemes (simple/complex and classification by afferent input) describe important differences among neurons in the cat's visual cortex and both must be incorporated into a comprehensive system for classifying cells in the cat's visual cortex; and (2) these two schemes when combined do not account for all the major differences among neurons in cat visual cortex, and at least two additional components are required: one related to orientation selectivity and dominance by the ipsilateral eye, and another related to directional selectivity and the degree of binocular interaction.