Towards non-parametric gridding of microarray images

cDNA microarrays, or biochips, are a technology used to measure gene expression on a large-scale basis. One of the critical issues of microarray experiments is the analysis of the produced images, which are the raw data from which measurements need to be made. However, the first stage, i.e. spot gridding, is not performed fully automatically in most microarray image analysis software. A novel, nonparametric gridding technique is proposed in this paper. Basically, the vertical and horizontal image projections are processed independently. An image is formed for each projection by plotting it into a 2D image, yielding a binary shape. Scale-space skeletonization is then performed in order to extract hierarchical representations of the signal. Using a criterion based on the number of blocks (or spots) in the microarray image, we select the scale in which the blocks(or spots) are detected. Experimental results for block segmentation, which constitutes the most difficult task in microarray gridding, are also shown.