Histopathological image classification using stain component features on a pLSA model

Abstract

Semantic annotation of microscopical field of views is one of the key problems in computer assistance of histopathological images. In this paper a new method for extracting patch descriptors is proposed and evaluated using a probabilistic latent semantic analysis (pLSA) classification model. The proposed approach is based on the analysis of the different dyes used to stain the histological sample. This analysis allows to find local regions that correspond to cells in the image, which are then described by the SIFT descriptors of the stain components. The proposed approach outperforms the conventional sampling and description strategies, proposed in the literature. © 2010 Springer-Verlag.