Automatic Breast Cancer Diagnostics Based on Statistical Analysis of Shape and Texture Features of Individual Cell Nuclei

Abstract

The automatic detection of nuclei within cytological samples is crucial for quantitative analysis in medical applications. Fortunately, modern digital microscopy systems allow imaging of biological material with very high accuracy. A typical cytological sample contains hundreds or thousands of cell nuclei that need to be examined for a particular type of cancer (or the exclusion of neoplastic lesions). Typically, this assessment is made by a qualified physician by visually analyzing a biological material. As the complexity of cellular structures is very high, automating this process is a big challenge. In this paper, we try to face this problem. Real cytological images of breast cancer patients were collected by pathologists from the University Hospital in Zielona Góra, Poland. The individual cell nuclei were automatically detected within cytological sample imagery. Then a couple of different shape and texture features were collected. Based on this data, an attempt was made to classify them in the context of the possibility of automatically identifying the type of cancer (malignant, benign). The results obtained are moderately promising.