Resumen
Summary. In an attempt to evaluate local architectural changes in oral epithelial premalignancy and malignancy, a quantitative method to analyse spatial cell arrangement as observed in 2D histological sections was developed based on mathematical morphology and graph theory. In total, 441 images (x20) of oral epithelium belonging to three diagnostic classes of interest (normal, dysplastic and neoplastic lesions) were assembled into collages for analysis. Epithelial cell nuclei markers were created from Haematoxylin and Eosin stained sections using colour deconvolution and morphological greyscale reconstructions. The epithelial tissue compartment was partitioned (using a digital watershed algorithm) into exclusive domains according to nuclei positions to approach the theoretical cell extents. The spatial arrangement of these 'cells' was then analysed in circular neighbourhoods of two sizes where four types of constrained graph networks (minimal spanning tree, relative neighbour graph, Gabriel graph and Delaunay triangulation) were constructed over the cell centroids. From these networks a total of 29 statistical properties were recorded. The statistical analysis of the network data indicated that unbiased and reproducible quantification of tissue architectural features is feasible and may provide valuable morphological information for diagnostic purposes and tissue characterization. (AU)
