Prediction of malignant transformation in oral epithelial lesions by image cytometry

The value of image analysis in predicting the malignant potential of oral epithelial lesions showing either hyperplasia or dysplasia was investigated; 5-micron formalin-fixed sections of 16 oral epithelial lesions, of which eight had later transformed to carcinoma and eight had not transformed during a follow-up of 10-15 years were studied. The sections were stained with the azure A-Feulgen reaction for nuclear DNA. In each section 200 nuclei of epithelial cells and 20 nuclei of lymphocytes were assessed; all measurements were made blindly. For each nucleus six features related to shape and amount of stain and six features related to chromatin pattern were assessed. For each feature the mean, SD, and interquartile range were determined and used for linear stepwise discriminant analysis. A model of three variables with the most discriminating power was developed. When the jackknifed classification test was applied using this model, the malignant potential of the lesions that later transformed could be predicted with 87.5% accuracy.