Abstract
A Quantimet 720D Image Analysis System has been programmed for light microscopic evaluation of the nuclear estrogen receptor distribution in frozen sections of human breast cancer stained by the peroxidase-antiperoxidase method using monoclonal antibodies to estrogen receptor protein (ER). This method provides precise criteria for distinguishing ER-positive and -negative cells and a sensitive and reproducible means for densitometric quantification of the staining patterns. Although imaging sequence and graphic analysis are automated by computer programs, light pen interaction provides supervision of feature selection. Imaging of the immunocytochemical assay (ER-ICA) in 50 patients revealed marked heterogeneity of nuclear estrogen receptor concentration varying over a nearly 100 fold concentration range. Various ER concentration patterns were evident: (I) distributions with a single peak (CV = 5%) present at various concentration levels; (II) bimodal distributions, revealing co-existent ER-positive and ER-negative subpopulations; (III) multimodal distributions with a number of resolvable concentration peaks; and (IV) highly skewed distributions with or without discernible peaks, frequently extending over the entire concentration range. Statistical methods of de-convolution were applied to determine the frequency and ER concentration characteristics of component subpopulations in the mosaic cases and for resolving the proportion of ER-positive and -negative cells. An approach for evaluating nuclear ER content in conjunction with ER concentration patterns in individual patients revealed whether spread in the ER concentration distribution resulted from differences in nuclear ER content or from variability in nuclear volume distribution.