A Unified Bayesian MRF-Based Poissonian Deconvolution And Segmentation Algorithm For Quantitative Colocalization Analysis In Dual-Color Fluorescence Microscopy

摘要

Colocalization studies perform dual-color fluorescence microscopy imaging of two (or more) biological entities in the same specimen to elucidate common functional characteristics from their spatial co-distribution. Reliable estimation of colocalization is challenging due to the presence of both noise and blur artifacts, in addition to the fluorescence intensity variations and common background, from the digital imaging process. State-of-the-art methods that quantify colocalization either require the input images to be preprocessed or fall short of addressing one or more of these challenges in a holistic way, thereby producing incorrect estimates. In contrast, this paper proposes a unified statistical framework to estimate colocalization using (i) a Bayesian Markov random field (MRF) modeling of the observed dual-channel image that is corrupted by (known) blur and Poisson noise, and (ii) the expectation-maximization (EM) algorithm for maximum-a-posteriori estimation of the MRF model parameters, including the degree of colocalization, and the restored intensity and segmented images. Experiments on several benchmark and laboratory datasets show that our method provides reliable estimates of colocalization and the underlying images over the state of the art, both qualitatively and quantitatively.