Premium
Architectural analysis of oral cancer, dysplastic, and normal epithelia
Author(s) -
Landini Gabriel,
Othman Ibrahim E.
Publication year - 2004
Publication title -
cytometry part a
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.316
H-Index - 90
eISSN - 1552-4930
pISSN - 1552-4922
DOI - 10.1002/cyto.a.20082
Subject(s) - pathology , linear discriminant analysis , graph , pattern recognition (psychology) , biology , medicine , artificial intelligence , mathematics , computer science , combinatorics
Background We present a novel, automated, and quantitative approach to evaluate local epithelial tissue architecture based on mathematical graph theory. Methods Four hundred forty‐one images of three diagnostic classes of oral epithelium (normal, dysplastic, and neoplastic) were analysed. The epithelial compartment was partitioned into exclusive areas associated with each nucleus to approach the theoretical cell extents. The spatial arrangement of cells in neighbourhoods of two sizes was characterised by constructing graph networks based on the cell centroids and recording 29 statistical properties. We analysed 104,627 and 67,590 neighbourhoods of diameters 37.5 and 75 μm, respectively. Results The discrimination power of the architectural descriptors was evaluated by using discriminant analysis. The best neighbourhood discrimination rate was 75% for normal versus carcinoma. For the pooled data, discrimination into three classes based on largest number of neighbourhoods associated with each class was 100% correct. Case‐wise, discrimination rates were 67%, 100%, and 80% correct for normal, premalignant, and malignant. When considering two classes, discrimination rates was 89% (normal) and 100% (malignant) correct, with 71% of premalignant cases assigned to the malignant class. Conclusions The results indicate that unbiased and reproducible quantification of tissue architectural features is possible and may provide valuable morphological information for diagnostic purposes. © 2004 Wiley‐Liss, Inc.