ABSTRACT
OBJECTIVE: To investigate the relevance of image analysis for grading breast carcinomas. STUDY DESIGN: The results of histologic grading were correlated with 18 features of image analysis, including SD. "Simple" characteristics, like area and perimeter, shape indices, optical density and textural features of nuclei from cancer cells, were analyzed. Hematoxylin-eosin-stained tissue sections of 67 cancer specimens were routinely used for the study. RESULTS: We found statistically significant correlations between overall histologic grading and the sum of its subscores and features of image analysis, especially nuclear area, nuclear perimeter and the diameter of the circumscribing circle (diametercirc), including their SDs. The visually and therefore subjectively assessed subscore of the nuclear pleomorphism of histologic grading significantly correlated with the features of image analysis, like nuclear area, nuclear perimeter, diametercirc, integrated optical density and correlation (and their SDs). There were significant relationships between the absolute numbers of mitoses per 10 high-power fields and nuclear area, nuclear perimeter and diametercirc (and their SDs). We did not observe a significant correlation between the subscore of tubule formation of histologic grading and any of the features of the image analysis studied. Furthermore, the correlations between the features of image analysis and the subscores of the visual histologic grading system were analyzed with respect to each other. The subscore of nuclear pleomorphism of histologic grading correlated best with overall grading (r = .72), whereas no significant correlation could be found between the subscores of nuclear pleomorphism and mitotic activity. CONCLUSION: Image analysis provides objectivity and reproducibility to the grading of breast carcinomas and thus could contribute to more individualized prognostication of the disease.
