Feature selection for segmentation of 2-D electrophoresis gel images

Abstract

Two-dimensional gel electrophoresis is the powerful technique used by biochemists to resolve and visualize protein samples. Commonly gels produced from several samples are analyzed in order to detect changes of protein expression. Thus computer-aided gel image analysis for protein spot detection became the main step in the whole process. Nevertheless accurate automatic spot detection is still difficult due to large variations in spot shape, image background and various inevitable artifacts. In this paper we investigate features of two-dimensional electrophoresis gel images. We look for those image features that will yield good results of protein spot detection done by the Feedforward Multilayer Neural Network. Feature comparison and spot segmentation results are presented and indicate that rotational symmetry features empowers segmentation of saturated and overlapped protein spots.