Centromere and cytoplasmic staining pattern recognition: a local approach

Autoimmune diseases are very serious and also invalidating illnesses. The benchmark procedure for their diagnosis is the indirect immunofluorescence (IIF) assay performed on the HEp-2 substrate. Medical doctors first determine the fluorescence intensity exhibited by HEp-2 wells and then report the staining pattern. Despite its pivotal role, IIF is affected by inter- and intra-laboratory variabilities demanding for the development of computer-aided-diagnosis tools supporting medical doctor decisions. With reference to staining pattern recognition, state-of-the-art approaches recognize five main patterns characterized by well-defined cell edges. These approaches are based on cell segmentation, a task that recent work suggests to be harder than the classification itself. In this paper, we extend the panel of detectable HEp-2 staining patterns, introducing the recognition of centromere and cytoplasmic patterns, which have a high specific match with certain autoimmune diseases, from other stainings. Since image segmentation algorithms fail on these samples, we developed a classification system integrating local descriptors and the bag of visual word approach, which represents image contents without the burden of segmentation. We tested our approach on a large dataset of HEp-2 images with high variability in both fluorescence intensity and staining patterns correctly recognizing the 97.12 % of samples. The system has also been validated in a daily routine fashion on 108 consecutive IIF analyses of hospital outpatients and inpatients, achieving an accuracy rate of 97.22 %.