Shape irregularity based area measurement: Application on cell nucleus recognition

Cell nucleus characterization is essential for pathological image analysis. In this article, we propose a shape characterization of the cell nucleus based on some selected shape descriptors. In order to separate automatically the different cell nucleus shapes, we focus our effort on the feature extraction and classification processes. The main contribution of our work is the creation of a new shape descriptor in order to evaluate the irregularity of the cell nucleus. This index takes its values between 0 and 1, and picks the value of 0 if and only if the measured shape is regular (circle, square, rectangle, triangle and hexagon). After segmentation step, three features are selected: elongation, circularity and irregularity factors. These parameters are used as input for ANFIS classifier when a set of rules are generated. The proposed method was evaluated with a data set of 40 nucleus. We obtain an accuracy of 97.62% with 3K-cross validation method.