Cytologic diagnosis of non-oxyphil follicular neoplasias of the thyroid. Multivariate classification procedure based...

The study was done on cytologic material of 58 non-oxyphil follicular neoplasias of the thyroid, 32 of which were adenomas and 26 carcinomas. Three groups of nucleolar features were quantified using a routine microscope with an ocular micrometer: frequency-, size-, and margination-related features. Since value overlap was present between two categories for all the variables, stepwise discriminant analysis was applied. The following three features were selected by the computer for calculation of one canonical discriminant function: percentage of marginated nucleoli, percentage of nuclei with one nucleolus, mean major nucleolar diameter. The percentage of agreement between morphologic and computer classifications was 95%. Two follicular adenomas were allocated to the carcinoma category, whereas one follicular carcinoma was allocated to the adenoma category. Out of 58, 52 were diagnosed by the computer into one of the two diagnostic categories with a very high probability, i.e. P greater than 0.75, the remaining 6 being considered intermediate.