Automated Classification System for HEp-2 Cell Patterns

Human Epithelial Type-2 (HEp-2) cells are essential in diagnosing autoimmune diseases. Indirect immunofluorescence (IIF) imaging is a fundamental technique for detecting antinuclear antibodies in HEp-2 cells. The four main patterns of HEp-2 cells that are being identified are nucleolar, homogeneous, speckled and centromere. The most commonly used method to classify the patterns is manual evaluation. This method is prone to human error. This paper will propose an automated method of classifying HEp-2 cells patterns. The first stage is image enhancement using Histogram equalization contrast adjustment and Wiener Filter. The second stage uses Sobel Filter and Mean Filter for segmentation. The third stage feature extraction based on shape properties data extraction. The last stage uses classification based on different properties data abstracted. The results obtained are more than 90% for nucleolar and centromere and about 70% for homogenous and speckled. For future work, another feature extraction method need to be introduced to increase the accuracy of the classification result. The method suggested is to analyze and obtain the data based on the texture of the image.