Multivariate karyometric approach in differential diagnosis of follicular thyroid neoplasms. A study of 31 cases.

Abstract

A retrospective analysis of 19 follicular adenomas, 12 minimally invasive follicular carcinomas and 3 widely invasive follicular carcinomas of the thyroid was performed on 5-microm-thick Feulgen-stained paraffin sections by means of a semiautomatic system for picture analysis. The major aim was to assess the potential of multiparameter karyometry for separation of the first two tumour types. Sixteen planimetric and densitometric features were defined in each case on 200-300 randomly selected nuclei and processed by a number of uni- and multivariate statistical methods. Despite predominantly significant ANOVA results a substantial overlap between tumour groups limited the practical usefulness of any karyometric feature alone. Factor and cluster analyses indicated independence of planimetric and densitometric parameters from each other, which was of crucial importance in finding an optimal subset of variables for discriminant analysis. The classification rule derived from the latter procedure was checked by the "jack-knife" method, by classification of 3 widely invasive cancers and by hierarchical tumour clustering. Sensitivity and specificity of the model for detection of malignancy were 100% and 94.7%, respectively. A multivariate karyometric approach, when applied correctly, can be a useful tool for differentiation between follicular adenomas and minimally invasive follicular carcinomas of the thyroid.