Automated microscope image analysis in the cytological tumor diagnostics. Results on borderline-cases of mammary...

50 fine-needle biopsies of mammary tumors (20 medullary carcinomas, 20 fibroadenomas, and 10 false negative judged cases) were studied by automated microscopic image analysis. Eight morphometric and densitometric features of the tumor cell nuclei were determined on Papanicolaou stained smears. It was analyzed whether smears false judged as negative can be rightly allocated as carcinomas by the automated method. The results of this study have shown that it was possible by means of three cell nucleus parameters (standard deviation of the area, mean area ratio, and skewness of the grey value gradient at contours) with 100% correct separation of the unequivocal cytologies into malignant and benign smears to recognize about 90% of the false judged smears rightly as carcinomas. The results are reproducible and can be obtained in a time of 2-5 min per smear. The first results show that the automated microscope image analysis may be a valuable tool in the diagnosis of borderline-cases of mammary tumors.