Objective: To identify quantitative histological features that can differentiate between low grade ductal carcinoma in situ (LG-DCIS) and classic type lobular carcinoma in situ (C-LCIS).
Study design: Regions of solid intraductal proliferation from scanned hematoxylin and eosin images of LG-DCIS (20 cases) and C-LCIS (25 cases) were analyzed using Mercator Version 1.0 (ExploraNova, La Rochelle, France). The system detects structures based on optical density. Multiple variables were measured, including estimators of size (mean nuclear area), shape (form factor), and staining (grayscale value). We also calculated the percentage that nuclei, cytoplasm, and white space represent in the total gland surface area.
Results: C-LCIS showed higher mean nuclear area, higher percentage of nuclear area, higher nuclear and cytoplasmic grayscale values, and lower percentage of cytoplasm area as compared to LG-DCIS. Combination of these variables using a random forest learning algorithm classified correctly 80% of LG-DCIS cases and 69% of C-LCIS cases.
Conclusion: Morphometric analysis is potentially a useful ancillary tool in the distinction between low-grade ductal and lobular in situ proliferations of the breast.