Independent Component Analysis of Optical Recordings from the.pdf

5th Joint Symposium on Neural Computation Proceedings 17 UCSD (1998) Independent Component Analysis of Optical Recordings from the Seaslug Tritonia Glen D. Brown, Satoshi Yamada, Hendrik Luebben, Terry Sejnowski ComputationalNeurobiology Laboratory, The Salk Institute for Biological Studies La Jolla, CA 92037 {glen,hendrik, terry}@salk.edu,yamada@bio.crl.melco.co.jp Abstract Independent component analysis (ICA) was used to separate action potential trains recorded with voltage-sensitive dyes in the isolated brain preparation of the seaslug Tritonia.It was assumed that the membrane potential of each neuron was an approximately independent source and that detectors (photodiodes) recorded linear mixtures of sources. These assumptions appear to be reasonable as ICA outperformed existing methods for separating spike trains in optical recording data. Spike trains, artifacts, and noise were assigned to different output channels (independent components). The ICA analysis method was also less labor intensive than other methods. Introduction Experimental approaches to the study of population coding require the collection and analysis of action potential trains from many neurons simultaneously.This may be accomplished with arrays of microelectrodes (Kruger, 1983;Wilson and McNaughton, 1993;Nordhausen et al. 1996;Nicolelis et al. 1997)or with voltage-sensitive dyes (Cohen and Lesher, 1986;Nakashirna et al. 1992).One problem faced by researchers recording populations of neurons is the assignment of actionpotentials to individual neurons. In gen