Staining Pattern Classification of Antinuclear Autoantibodies Based on Block Segmentation in Indirect Immunofluorescence Images

• Published in: PloS one Vol. 9; no. 12; p. e113132
• Main Authors: Li, Jiaqian, Tseng, Kuo-Kun, Hsieh, Zu Yi, Yang, Ching Wen, Huang, Huang-Nan
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 04.12.2014; Public Library of Science (PLoS)
• Subjects: Algorithms; Antibodies, Antinuclear - immunology; Antibodies, Antinuclear - isolation & purification; Antinuclear antibodies; Autoantibodies; Autoimmune diseases; Autoimmune Diseases - diagnosis; Autoimmune Diseases - immunology; Autoimmune Diseases - pathology; Autoimmunity; Biology and Life Sciences; Cell Fusion; Cell Tracking - methods; Classification; Classifiers; Cluster Analysis; Comparative analysis; Computer and Information Sciences; Computer science; Fluorescent antibody technique; Fluorescent Antibody Technique, Indirect - methods; Hep G2 Cells; Humans; Image classification; Image processing; Image Processing, Computer-Assisted; Image segmentation; Immunofluorescence; International conferences; Medicine and Health Sciences; Morphology; Neural networks; Optical Imaging - methods; Pattern classification; Pattern recognition; Principal components analysis; Staining; Substrates; China; Taiwan
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0113132

Indirect immunofluorescence based on HEp-2 cell substrate is the most commonly used staining method for antinuclear autoantibodies associated with different types of autoimmune pathologies. The aim of this paper is to design an automatic system to identify the staining patterns based on block segmentation compared to the cell segmentation most used in previous research. Various feature descriptors and classifiers are tested and compared in the classification of the staining pattern of blocks and it is found that the technique of the combination of the local binary pattern and the k-nearest neighbor algorithm achieve the best performance. Relying on the results of block pattern classification, experiments on the whole images show that classifier fusion rules are able to identify the staining patterns of the whole well (specimen image) with a total accuracy of about 94.62%.