Diagnosis of autoimmune diseases using HEp-2 staining pattern and local derivative pattern features

Indirect Immunofluorescence (IIF) of Human Epithelial Type-2 (HEp-2) cells treated with blood serum can determine the presence of various antinuclear antibodies. This leads to diagnosis of autoimmune diseases such as systemic lupus erythematosus, Sjogren's syndrome, and rheumatoid arthritis. Although it is reliable, the subjective nature of interpretation in manual IIF method gives rise to significant inter-observer variances. Hence, a computer aided system is necessary. In this work, an attempt has been made to classify HEp-2 staining patterns using Local Derivative pattern (LDP) features. Standard images from a public domain database are used in this study. It consists of 1500 HEp-2 cell images belonging to five classes of staining patterns, namely centromere, coarse speckled, fine speckled, homogenous, and nucleolar. The images are preprocessed to improve contrast using contrast stretching technique. Subsequently, nuclear particles are segmented using Otsu thresholding and validated against ground truth provided in the dataset. LDP features along with global features such as area, entropy and mean intensity are extracted from the segmented images. Subsequently, these features are used for the differentiation of texture patterns using SVM classifier with a polynomial kernel. The obtained results indicate that Otsu based thresholding after contrast stretching is able to segment nuclear particles in all the images. The extracted features including the LDP features prove to be useful in classifying HEp-2 staining pattern with an accuracy of 73.73%. As IIF microscopy using HEp-2 cells is the gold standard for autoimmune disease diagnosis, the proposed work seems to be highly relevant in a clinical setting.