HISTOMETRIC TEXTURE ANALYSIS OF DNA IN THIN-SECTIONS FROM BREAST BIOPSIES - APPLICATION TO THE DETECTION OF MALIGNANCY-ASSOCIATED CHANGES IN CARCINOMA IN-SITU

OBJECTIVE: To evaluate histometric measurement of nuclear texture in b reast biopsy sections in order to detect malignancy-associated changes in apparently normal tissue in the vicinity of carcinoma in situ. STU DY DESIGN: We previously showed that image cytometry measurements of n uclear features-foremost, texture features, describing the organizatio n of Feulgen-stained DNA; in the cell-can be used to distinguish norma l-appearing, diploid epithelial cells from patients with invasive carc inoma of the breast from those with benign biopsies. In that study, re ferred to as the ''single cell analysis,'' images of at least 200 epit helial cells were acquired for each slide, and substantial user intera ction was required to segment cells from each field. Location of isola ted cells and interactive segmentation are both time-consuming procedu res, particularly in breast tissue, where nuclei can be tightly cluste red within a duct. With histometric texture analysis on the same speci mens, segmentation of individual cells was ignored, and texture measur ements were performed over the entire cluster of relevant cells. With this approach, ploidy information is not available, and touching and o verlapping nuclei are included in the measurements. Measurement of his tometric texture properties requires substantially less time (at feast an order of magnitude) than individual cell measurement and, if ploid y information is not significant, may therefore provide a move practic al means of analysis for tissue sections. RESULTS: Seventeen cases of invasive carcinoma and 17 cases of nonproliferative breast disease wer e examined. Using stepwise discriminant function analysis, slides were classified into one of the two groups with an accuracy of 88.6% in th e case of single cell analysis and with an accuracy of 88.2% using his tometric analysis. CONCLUSION: The existence of malignancy-associated changes in the breast wits confirmed by an independent analysis of the same specimens. Although the two methods are not directly comparable, we found that histometric texture analysis performs at least as well as single-cell analysis for the detection of malignancy-associated cha nges in breast carcinoma.