CHEMOMETRIC LABELING OF CEREAL TISSUES IN MULTICHANNEL FLUORESCENCE MICROSCOPY IMAGES USING DISCRIMINANT-ANALYSIS

This paper presents a novel, semiautomatic method for microscopic iden tification of multicomponent samples, which allows the identification, location, and percentage quantity of each component to be determined. The method involves applying discriminant analysis to a sequence of m ultichannel fluorescence microscopy images via a supervised learning a pproach; by selecting groups of pixels that are representative for eac h component type in a ''known'' sample, a computer is ''taught'' how t o recognize the behavior (i.e., fluorescence emission) of the various components when illuminated under different spectral conditions, The i dentity, quantity, and location of these components in ''unknown'' sam ples (i.e., samples with the same component types but in different rat ios or distributions) can then be investigated. The technique therefor e enables semiautomatic quantitative fluorescence microscopy and has p otential as a quality control tool, This work demonstrates the applica tion of the technique to artificial and natural samples and critically discusses its quality, potential, and limitations.