An Automatic Classification Model for ANA HEP2-Cell Pattern Images

The Human immune system is responsible for protecting the human body from external threats. Autoimmune diseases are a class of diseases that occur when the immune system of a person behaves abnormally. A person with autoimmune disease produces Antinuclear Antibodies (ANA) instead of antibodies. The standard gold test to detect autoimmune diseases is detecting the concentration and localization of ANAs using the Immunofluorescence test (IFA) test. Human Epithelial (HEp2) cells were used as the substrate in the IFA test. Localization and specificity of these ANA on Human Epithelial (HEp2), which stains fluorescent color, gives the diagnosis the different autoimmune diseases. Automation is required in the detection and classification of these patterns to increase throughput in the pathology labs. In the current study different patterns produced by ANA on HEp2 cells is detected. The work mainly focused on the design of a generic classifier that classifies the images of different qualities. Current study images of different sizes and intensities are considered from the two databases, initially, used individual data base and combined them later. An attempt of a systematic study carried out from the handcrafted feature extraction to explore Machine learning classification followed by fine-tuning the parameter as mentioned above. Later CNN approach is also carried out. In which the CNN approach is considered as the better. ResNet50 scored better in comparison with VGG16 it has solved the reducing gradient issue. Grad-CAM is used to visualize the intermediate features layer of test images in CNN in ResNet50