Confocal Microscopy and Nuclear Segmentation Algorithm for Quantitative Imaging of Epithelial Tissue

Abstract

Carcinomas, cancers that originate in the epithelium, account for more than 80% of all cancers. When detected early, the 5-year survival rate is greatly increased. Biopsy and histopathology is the current gold standard for diagnosis of epithelial carcinomas which is an invasive, time-intensive, and stressful procedure. In vivo confocal microscopy has the potential to non-invasively image epithelial tissue in near-real time. This dissertation describes the development of a confocal microscope for imaging epithelial tissues and an image processing algorithm for segmentation of epithelial nuclei.

A rapid beam and stage scanning combination was used to acquire fluorescence confocal images of cellular and tissue features along the length of excised mouse colon. A single 1 × 60 mm2 field of view is acquired in 10 seconds. Disruption of crypt structure such as size, shape, and distribution is visualized in images of inflamed colon tissue, while the normal mouse colon exhibited uniform crypt structure and distribution.

An automated pulse coupled neural network segmentation algorithm was developed for epithelial nuclei segmentation. An increase in nuclear size and the nuclear-to-cytoplasmic ratio is a potential precursor to pre-cancer development. The spiking cortical model algorithm was evaluated using a developed confocal image model of epithelial tissues with varying contrast. It was further validated on reflectance confocal images of porcine and human oral tissue from two separate confocal imaging systems. Biopsies of human oral mucosa are used to determine the tissue and system effects on measurements of nuclear-to-cytoplasmic ratio.