Classification of potential nuclei in prostate histology images using shape manifold learning

The demanding step in the development of ancillary systems for the diagnosis of cancer and other diseases based on nuclear morphometry is the delineation of nuclei in the images of stained tissue sections. Various constituents of the tissue section such as cellular and extra-cellular elements, staining artefacts, debris of nuclei, and clusters of overlapping nuclei apart from the image acquisition noise to name a few contribute to in the complexity of the task. In this paper, we pose the problem of selection of nuclei in tissue section as classification of shapes using manifold learning on training images followed by out-of-sample extension for unknown test images. Experimental results demonstrate the effectiveness of the proposed algorithm.