Semi-Supervised Learning and Generalized Mixture Models in the analysis of Retinal Images

Samuel J. Frame
S. Rao Jammalamadaka
Department of Statistics and Applied Probability
University of California, Santa Barbara

Abstract

The Bio-Image Informatics research program at UCSB is a collaborative effort funded by the National Science Foundation. The statistical computing problem we address in this paper is to tailor an application of the Generalized Mixture Models (GMM's) for analyzing biological images, and the development of versatile software that is useful for this purpose. This analysis has several goals. First, we demonstrate that such methods can be used as objective diagnostic tools for classifying new images in the medical and biological context, instead of relying on subjective human analysis. Second, we are interested in using GMM's to better understand the similarities and differences between various classes of controlled experiments. We do this by inspecting and learning from fitted model components. Further, we are able to test for the equality of local regions (represented by model components) from different known classes. After a brief introduction of the GMM's, we discuss a case-study that has been of interest to biologists in this project.
[PDF] [BibTex]
Samuel J. Frame, S. Rao Jammalamadaka,
Center for Bio-Image Informatics, UCSB, Mar. 2006.
Node ID: 437 , DB ID: 240 , Lab: BIO , Target: Technical Report