Polar modelling and segmentation of genomic microarray spots using mathematical morphology

Abstract

Robust image analysis of spots in microarrays (quality control + spot segmentation + quantification) is a requirement for automated software which is of fundamental importance for a high-throughput analysis of genomics microarray-based data. This paper deals with the development of model-based image processing algorithms for qualifying/segmenting/ quantifying adaptively each spot according to its morphology. A series of morphological models for spot intensities are introduced. The spot typologies represent most of the possible qualitative cases identified from a large database (different routines, techniques, etc.). Then, based on these spot models, a classification framework has been developed. The spot feature extraction and classification (without segmenting) is based on converting the spot image to polar coordinates and, after computing the radial/angular projections, the calculation of granulometric curves and derived parameters from these projections. Spot contour segmentation can also be solved by working in polar coordinates, calculating the up/down minimal path, which is easily obtained with the generalized distance function. With this model-based technique, the segmentation can be regularised by controlling different elements of the algorithm. According to the spot typology (e.g., doughnut-like or egg-like spots), several minimal paths can be computed to obtain a multi-region segmentation. Moreover, this segmentation is more robust and sensible to weak spots, improving the previous approaches.