The digital revolution in cellular and sub-cellular microscopy has had a remarkable effect on biological research. High throughput imaging technology has enabled the rapid acquisition of massive volumes of biological images. Sophisticated imaging methods are producing biological images of unprecedented detail and resolution. These advances have facilitated the capture of entire tissues, cell populations and even organs that were unimaginable until recently. The potential for significant scientific discovery from these large and detailed images is enormous.

Yet many challenges remain in managing and mining these images in order to conduct biological inference. Biological images present unique computational problems that are rarely encountered in other image mining and management domains. Mining for patterns in tissue is often difficult due to the high density and heterogeneity of cells. Capturing biological processes from sensitive tissue at the limit of light resolution frequently produces noisy and poor quality images, resulting in a high degree of uncertainty of the underlying biology. These bio-image specific challenges have led to the emergence of a burgeoning new field, bio-image informatics, which aims to tackle the computational roadblocks to extracting biological knowledge from cellular and sub-cellular imaging.

In this dissertation, I present my research on managing and mining uncertain biological images. I propose a method to quantify the spatial correlation between large, continuous structures and point objects, such as those found in the retina. Furthermore, I detail a model of spatial network creation and its application in biological tissue. Next, I present a novel algorithm to significantly improve the performance of searching for similar images or probability distributions in a biological database. Finally, I present image analysis work flow that incorporates image uncertainty in order to extract biological knowledge from images. While all of these methods are directly applied to biological tissues and images, they have broad applications in many other domains.