Spectral morphometric characterization of breast carcinoma cells

The spectral morphometric characteristics of standard haematoxylin and eosin breast carcinoma specimens were evaluated by light microscopy combined with a spectral imaging system. Light intensity at each wavelength in the range of 450–800 nm was recorded for 10(4) pixels from each field and represented as transmitted light spectra. A library of six characteristic spectra served to scan the cells and reconstruct new images depicting the nuclear area occupied by each spectrum. Fifteen cases of infiltrating ductal carcinoma and six cases of lobular carcinoma were examined; nine of the infiltrating ductal carcinoma and three of the lobular carcinoma showed an in situ component. The spectral morphometric analysis revealed a correlation between specific patterns of spectra and different groups of breast carcinoma cells. The most consistent result was that lobular carcinoma cells of in situ and infiltrating components from all patients showed a similar spectral pattern, whereas ductal carcinoma cells displayed spectral variety. Comparison of the in situ and the infiltrating ductal solid, cribriform and comedo carcinoma cells from the same patient revealed a strong similarity of the spectral elements and their relative distribution in the nucleus. The spectrum designated as number 5 in the library incorporated more than 40% of the nuclear area in 74.08% of the infiltrating lobular cells and in 13.64% of the infiltrating ductal carcinoma cells (P < 0.001). Spectrum number 2 appeared in all infiltrating ductal cells examined and in none of the lobular cells. These results indicate that spectrum number 5 is related to infiltrating lobular carcinoma, whereas spectrum number 2 is characteristic for infiltrating ductal carcinoma cells. Spectral similarity mapping of central necrotic regions of comedo type in situ carcinoma revealed nuclear fragmentation into defined segments composed of highly condensed chromatin. We conclude that the spectral morphometric features found for lobular and ductal cell populations may serve future automated histological diagnostics. © 1999 Cancer Research Campaign