Mathematical morphology applied to spot segmentation and quantification of gene microarray images

Abstract

Abstract-DNA microarray technology is a very powerful technique used in modern biology which is extensively used for identification of sequence (gene/gene mutation) and determination of gene expression. A typical gene microarray image consists of a few hundred to several thousand spots and the extent of hybridization of these spots determines the level of gene expression (abundance) in the sample. The massive scale and variability of gene microarray data creates new challenging problems of gene clustering, feature extraction and data mining. A major issue in gene microarray data analysis is to accurately quantify spot shapes and intensities of microarray image. In this paper we propose a method for performing accurate spot segmentation of a microarray image, using morphological image analysis techniques, followed by quantification of the shapes of the segmented spots using B-Splines.

Clustering-based spot segmentation of cDNA microarray images

“…Fixed circle segmentation fixes diameters of circles characterizing the spots [11] whereas adaptive circle segmentation dynamically changes the diameter of circles for the spots [12]. Adaptive shape segmentation methods including watershed transform [13], seeded growing region [14,15] and Markov random field [16] are the other image processing approaches. Histogram based segmentation [17] calculates a threshold value where intensity values higher than this threshold are treated pixels of the spots.…”

Image Processing and Machine Learning Techniques for the Segmentation of cDNA Microarray Images

“…The pixels that belong to the watershed lines are assigned a special label. Siddiqui et al (Siddiqui et al, 2002) had implemented a segmentation method where the watershed transform is applied to the gradient of the image (Serra, 1982) and not to the original one. The gradient operator is very sensitive to grayscale variation and noise and it can cause development of a large number of irrelevant catchment basins.…”

“…Histogram-based techniques estimate a threshold (GSI Lumonics, 1999;Chen et al, 1997), such that pixels with intensity lower than the calculated threshold are characterized as background pixels, whereas pixels with higher intensity as signal pixels. The adaptive shape segmentation methods are usually based on the Watershed Transform (Siddiqui et al, 2002) and the Seed Region Growing algorithm (Buckley, 2000;Wang et al, 2001). The most recent techniques employ clustering algorithms such as K-means (Bozinov & Rahnenführer, 2002;Ergüt et al, 2003;Wu & Yan, 2004), Fuzzy C-Means (FCM) (Ergüt et al, 2003), Expectation-Maximization (EM) (Blekas et al, 2005) and Partitioning Around Medoid ( method (Rahnenführer & Bozinov, 2004) which engages Image Processing and Machine learning techniques has been proposed.…”

A Robust Algorithm for Ratio Estimation in Two-Color Microarray Experiments

“…There have been a number of publications reporting application of the median filter, 6,18 top-hat filter 4,6 or a set of morphological operators 3,19 to facilitate further image processing. However, all these techniques, while reducing noise in images, also change intensity levels of the majority of pixels on the array, regardless of whether these pixels are outliers or not.…”