Optical Imaging As Source Separation Problem

Abstract

If we want to reliably infer neuronal activity patterns from optical imaging data, we face two problems: (i) Most of the optical signals represent only an indirect measure of neuronal activity. This is particularly true for evoked metabolic changes such as intrinsic signals. Therefore, the relationship between neuronal activity and optical signals must be characterized in detail. The only direct method for doing this is the simultaneous electrophysiological and optical measurement of neuronal activity, which has been done for intrinsic signal imaging (Malonek and Grinvald, 1996). We will not further treat direct comparison, but will later discuss some indirect methods for estimating to what extent intrinsic signal components reflect neuronal activity, (ii) Only some of the measured changes in light reflectance are suitable for monitoring neuronal activity. They are mixed with unwanted components, which either reflect only the global activation level or are independent of neuronal activity. We need to separate the signal components in the image stack, which are closely and locally coupled to neuronal activity, from die remaining components. In other words, dye signals or intrinsic mapping signals must be reliably separated from global signals, blood vessel patterns, hemodynamics and ongoing activity in order to obtain a (still indirect) measure for two-dimensional neuronal activity distributions over space.