Classification of Multicolor Fluorescence in Situ Hybridization Images Using Gaussian Mixture Models

This paper introduces a fully automatic chromosome classification algorithm for Multicolor or Multiplex Fluorescence In-Situ Hybridization (M-FISH) images using Gaussian mixture model technique. M-FISH is a recently developed cellular imaging method for rapid detection of chromosomal abnormalities, where each chromosome is labeled with 5 dyes and counterstained with DAPI florescence stain. The problem is modeled as a 24class 6-feature pixel-by-pixel classification task. Features are composed of dyed-image pixel brightness. By conducting experiments on ADIR M-FISH database, we demonstrate that our proposed Gaussian mixture model based approach yields significantly better results over the Bayesian classifiers which assume a single Gaussian probability distribution for M-FISH data.