HEP-2 CELL IMAGES CLASSIFICATION BASED ON STATISTICAL TEXTURE ANALYSIS AND FUZZY LOGIC

Autoimmune diseases occur when an inappropriate immune response takes place and produces autoantibodies to fight against human antigens. In order to detect autoimmune disease, a test called indirect immunofluorescence (IIF) will be carried out to identify antinuclear autoantibodies (ANA) in the HEp-2 cell. The outcome of the test includes observing fluorescence intensity of the sample and classifying the staining pattern of the cell. Current method of analysing the results is limited to subjective factors such as experience and skill of the medical experts. The results obtained from the visual analysis are debatable as it is inconsistent. Thus, there is a need for an automated recognition system to reduce the variability and increase the reliability of the test results. Automated system also saves time and cost as the system is able to process large amount of image data at one time. This project proposes a pattern recognition algorithm consisting of statistical methods to extract seven textural features from the HEp-2 cell images followed by classification of staining patterns by using fuzzy logic. This method is applied to the data set of the ICPR 2012 contest in which each cell has been manually segmented and annotated by specialists. The textural features extracted are based on the first-order statistics and second-order statistics computed from grey level co-occurrence matrices (GLCM). The first-order statistics features are mean, standard deviation and entropy while the features extracted by GLCM are contrast, correlation, energy and homogeneity. The extracted features will then be used as an input parameter to classify the staining pattern of the HEp-2 cell images by using Fuzzy Logic. The staining patterns are divided into five categories; homogeneous, nucleolar, centromere, fine speckled and coarse speckled. A working classification algorithm is developed by using MATLAB and the Fuzzy Logic Toolbox to differentiate and classify the staining pattern of HEp-2 cell images. The algorithm gives a mean accuracy of 84% out of 125 test images.