Semantic Segmentation of Microscopic Images Using a Morphological Hierarchy

Abstract

The objective of semantic segmentation in microscopic images is to extract the cellular, nuclear or tissue components. This problem is challenging due to the large variations of these components features (size, shape, orientation or texture). In this paper we present an automatic technique to robustly identify the epithelial nuclei (crypt) against interstitial nuclei in microscopic images taken from colon tissues. The relationship between the histological structures (epithelial layer, lumen and stroma) and the ring like shape of the crypt are considered. The crypt inner boundary is detected using a closing morphological hierarchy and its associated binary hierarchy. The outer border is determined by the epithelial nuclei, overlapped by the maximal isoline of the inner boundary. The evaluation of the proposed method is made by computing the percentage of the mis-segmented nuclei against epithelial nuclei per crypt.