Analysis of HEp-2 images using MD-LBP and MAD-bagging

Abstract

Indirect immunofluorescence imaging is employed to identify antinuclear antibodies in HEp-2 cells which founds the basis for diagnosing autoimmune diseases and other important pathological conditions involving the immune system. Six categories of HEp-2 cells are generally considered, namely homogeneous, fine speckled, coarse speckled, nucleolar, cyto-plasmic, and centromere cells. Typically, this categorisation is performed manually by an expert and is hence both time consuming and subjective. In this paper, we present a method for automatically classifiying HEp-2 cells using texture information in conjunction with a suitable classification system. In particular, we extract multidimensional local binary pattern (MD-LBP) texture features to characterise the cell area. These then form the input for a classification stage, for which we employ a margin distribution based bagging pruning (MAD-Bagging) classifier ensemble. We evaluate our algorithm on the ICPR 2012 HEp-2 contest benchmark dataset, and demonstrate it to give excellent performance, superior to all algorithms that were entered in the competition.