Exploring Spatial Information for Improved Microarray Image Segmentation and Quality Assessment

• Activity Number: 510
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Minyao Sun*+ and Yan Yang
• Companies: Arizona State University and Arizona State University
• Keywords: artifact detection ; connected-component labeling ; EM algorithm ; mixture model ; saturation

Abstract

Microarray technology has been widely used in biomedical research to study the function of tens of thousands of genes in a single experiment. Prior to downstream analysis (e.g., identification of differentially regulated genes) and the biological interpretation of microarray data, appropriate methods for extracting the hybridization signals from microarray images are essential. However, artifacts caused by dust, fibers, scratches, etc. that are not uncommon on an array can seriously contaminate the signals, whereas tools for automatic quality assessment are lacking. We exploit the spatial information to guide model-based segmentation and spot intensity estimation. Also, we utilize the spatial as well as spot shape information to develop quality assessment measures for detecting potential artifacts. A segmentation method is developed that iterates between mixture model-based clustering and a four-neighbor connected component (FNCC) labeling algorithm to reduce the distorting effect of small disconnected artifacts on cluster memberships. Spot-level quality assessment measures are formulated to automatically detect artifacts of various shapes and sizes. The affected spots are then flagged for downstream analysis. The proposed methods are illustrated with the biomedical data from a human Valley Fever diagnosis study. Our segmentation procedure produces spatially connected clusters free of small bright artifacts. The quality control tools developed can be used to identify dust and fibers that typically expand several spots and are potentially useful for detecting arrays contaminated by scratches. In addition, poor gridding and non-normal spot types (e.g., blank spots, black holes, donut-shaped spots, and irregularly-shaped spots with weak signal) are automatically identified, providing additional information on local and global quality control.