Selection of relevant texture descriptors for recognition of HEp-2 cell staining patterns

One of the important applications of computer-aided diagnosis is the detection of connective tissue disorders through automatic classification of antinuclear autoantibodies (ANAs). The recognition of ANAs is primarily done by analyzing indirect immunofluorescence (IIF) images of human epithelial type 2 (HEp-2) cells. In this regard, the paper introduces a novel approach for automatic classification of ANAs by staining pattern recognition of HEp-2 cell IIF images. Considering a set of HEp-2 cell images, the proposed method selects a set of relevant local texture descriptors for a pair of staining pattern classes, as well as identifies a set of important features corresponding to each relevant descriptor. The set of features for multiple classes is obtained from each of the important feature sets selected under various relevant local texture descriptors for all possible class-pairs. The relevance of a descriptor is evaluated based on the theory of rough hypercuboid approach, while the important feature set of a local descriptor is formed by reducing the impact of both noisy pixels present in an HEp-2 cell image and noisy HEp-2 cell images in a staining pattern class. Finally, the support vector machine is used to recognize one of the known staining patterns present in IIF images. The effectiveness of the proposed staining pattern recognition method, along with a comparison with related approaches, is illustrated on two benchmark databases of HEp-2 cell images using different classifiers and experimental set-up. The results show that the proposed approach performs significantly better than existing methods, with respect to both classification accuracy and F1 score, irrespective of the databases and classifiers used.