Probabilistically assisted spot segmentation — With application to DNA microarray images

Abstract

Based on the assumption that a microarray spot has an approximately circular shape we here introduce a robust method for detection of spots located inside defined regions (“spot boxes”) of digital images. By appropriate vectorization of the spot box pixel positions, the spot detection is completed by maximization of a two-sample t-statistic. The method presented outperforms established segmentation methods on a benchmark dataset of various artificially generated DNA microarray images. The method is also demonstrated to be superior to the strongest competitors (as identified by processing the artificial data) when compared on a real set of microarray images.