Unsupervised segmentation of cell nuclei using geometric models

Fluorescent microscopy of biological samples allows non-invasive screening of specific molecular events in-situ. This approach is useful for investigating intricate signalling pathways and in the drug discovery process. The large volumes of data involved in image analysis are a limiting factor. As manual image interpretation relies on expensive manpower automated analysis is a far more appropriate solution. In this paper we discuss our approach to achieve reliable automated segmentation of individual cell nuclei from wide field images taken of prostate cancer cells. We present a novel analysis routine to accurately identify cell nuclei based upon intensity clustering and morphological validation using a data derived geometric model. This approach is shown to consistently outperform the standard analysis technique using real data.