Statistical analysis and modeling of biological fluorescence images. Methods and applications
Abstract
In fluorescence microscopy, the accurate measurement of spatial closeness (colocalization) between two (or more) fluorescent labels each having a different wavelength, is a hotly disputed topic. The colocalization interpretation through image analysis-based techniques is often uncertain because image pairs are mostly analyzed based on a pixel-to-pixel comparison and the correlation between the images is usually taken as a measure of colocalization. Widely used colocalization measures such as Pearson's correlation coefficient (PCC), Mander's overlap coefficients (MOC), and intensity correlation quotient (ICQ) calculate colocalization by treating each pixel independently and therefore fail to determine whether the observed value occurred by chance and what will happen if two markers are not overlapping but still close enough to interact? Additionally, owing to the limited photon budget nature of fluorescence images, the correlation-based colocalization quantification is severely dependent on the image quality. To address this, we developed a robust methodology based on the doubly stochastic Gauss-Poisson Markov random field model for colocalization quantification. Our approach compares neighborhoods is defined by an Markov random field (MRF) and not by individual pixels; thereby giving accurate colocalization values even if the images are uncorrelated in strict mathematical sense. Our method partitions the pair of images to be tested for colocalization into statistically uniform segments (superpixels), the superpixels are subsequently treated as a realization of a doubly stochastic random field, and the parameters of the underlying random field are estimated for every superpixel. The estimated random fields are compared instead of pixels to measure the spatial overlap. The robustness of the proposed method to fluorescent surges make reliable colocalization measurements independent of illumination intensity and camera settings.
Degree
Ph.D.
Advisors
Talavage, Purdue University.
Subject Area
Biostatistics|Engineering|Biomedical engineering|Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.