Rough-Bayesian approach to select class-pair specific descriptors for HEp-2 cell staining pattern recognition

Abstract

One of the important problems in computer-aided diagnosis of connective tissue disease is automatic recognition of staining patterns present in HEp-2 cells. In this regard, the paper introduces a novel approach for the recognition of staining patterns by HEp-2 cell indirect immunofluorescence image analysis. The proposed method assumes that a fixed set of local texture descriptors or scales may not be effective for classifying staining patterns into multiple classes. A particular set of descriptors or scales may be significant for classifying a pair of classes, but may not be relevant for other pairs of classes. The proposed approach, therefore, first selects a set of local texture descriptors under appropriate scales for each class-pair, and then forms the final feature set for multiple classes from the relevant descriptors of all possible pairs of classes. A novel framework, termed as Rough-Bayesian model, is introduced to evaluate the relevance of a descriptor and/or a scale. It is based on the merits of rough sets and Bayes decision theory. During the selection of relevant descriptor and/or scale, the proposed method takes care of the presence of both noisy pixels in an HEp-2 cell image and noisy HEp-2 cell images in a staining pattern class. The support vector machine is used to predict the staining patterns present in HEp-2 cell images. The performance of the proposed method, along with a comparison with state-of-the-art methods, is demonstrated on several HEp-2 cell image databases. An important finding is that the accuracy for classifying HEp-2 cell images is significantly increased if class-pair specific descriptors under appropriate scales are considered, instead of selecting a uniform set of descriptors and scales for multiple classes.