Results of the use of Bayesian classifiers for identification of breast cancer cell nuclei

The discrimination between breast carcinoma cell nuclei and artifacts from cytological images is one key element for achieving analysis of cell characteristics, which is useful for tumour prognosis. In this paper the identification of cell nuclei is carried out by means of a Bayesian classifier integrating reject options and its use for classes whose distributions have unequal covariance matrices is analysed. This analysis is more general than the one introduced earlier by Dubuisson and Masson (1992), and shows that there are other consequences from using rejects to be discussed. We make a comparison between two Bayesian classifiers and the experimental results are presented