Classification of normal colorectal mucosa and adenocarcinoma by morphometry

Semi-automatic image analysis was used to make a morphometrical assessment of 15 nuclear and cellular variables in normal (n = 20) and malignant (n = 30) colorectal epithelium. Principal components analysis on the matrix of correlations between variables identified four main sources of variation within the dataset. These were, in decreasing order of importance: (1) nuclear size, nuclear cytoplasmic ratio and nuclear position within the cell; (2) the variability of nuclear size; (3) nuclear elongation and polarity; (4) nuclear shape and its variation. Discriminant analysis was conducted between histologically normal mucosa (n = 10) and adenocarcinoma in ulcerative colitis (n = 20). Using stepwise variable selection, the mean nuclear cytoplasmic ratio (normal, mean 20.4 (s.d. +/- 2.0); tumour, mean 39.7 (s.d. +/- 7.0)) and the coefficient of variation of nucleus to cell apex distance (normal, mean 19.2 (s.d. +/- 7.5); tumour, mean 47.8 (s.d. +/- 9.1)) were chosen as discriminating features. They were used to derive a discriminant function which gave perfect discrimination between the two groups. Scatter plots of these two variables confirmed complete separation of normal mucosa from adenocarcinoma and provided a simple method of applying the discriminant function. Discriminatory performance did not deteriorate when the function was applied to further normals (n = 10) and adenocarcinoma (n = 10). This study highlights the descriptive differences between normal and malignant colorectal epithelium and shows that case allocation may be made to these two lesion categories using a morphometrically-derived classification rule.