På svenska

Presentation Information     2007-09-10 (15:15)   •  The seminar room at Vi2

Speaker Milan Gavrilovic
Title Algorithms for Detection of Colocalization Based on Spectral Decomposition
Abstract Most of the popular methods for detection of colocalization are based on intensity thresholding of a 2D histogram which correlates the relationship between two fluorescence stains, providing segmentation and localization of colocalization at the same time. A novel set of algorithms is proposed, which classify image pixels as colocalized or not based on angles, or ratios of intensities rather than individual intensities, within a 2D histogram. Intensity thresholding may thereafter be used as a post-processing step to suppress background noise. This separation of pixel classification and image segmentation allows for the use of any type of advanced segmentation in combination with quantification of colocalization. Also, basing pixel classification on angles rather than pure intensities provides automated compensation (and quantification) of cross-talk and bleed-through between fluorescence stains. The proposed algorithms are compared to well known and commonly used methods for quantification of colocalization and have proven to be more stable and robust for a number of artificial and real image situations.