A statistical model for fluorescence image cytometry

Abstract

Fluorescence image cytometry is a common laboratory method used to analyze tissue and culture specimens at the cellular level. Fluorescence imaging is useful because fluorescent stains are highly specific and imaging allows for direct spatial measurements. A statistical model was developed for analysis of fluorescence images. The model incorporates a nonparametric specification for the characteristic cell shape in the image. A realistic representation of the image data is achieved by flexible positioning, orientation and rescaling of each cell and summation of the contributions from a number of distinct cells. Advantages of this approach relative to current methods based on image segmentation include the improved ability to distinguish clustered structures and the ability to naturally incorporate blurring. The method of regularization is used to estimate the parameters of the model. Simulation studies show that the method is consistent for all parameters and gives reasonable estimates even in the presence of substantial image noise. The method is illustrated with data from a Tangier disease experiment.