HEp-2 Cell Image Classification Based on Convolutional Neural Networks

Autoimmune diseases are the third cause of mortality in the world. A conventional method to support the diagnosis of Autoimmune diseases is the identification of anti-nuclear antibody (ANA) via Immunofluorescence (IIF) test in human epithelial type-2 cells (HEp-2). In the present work, a new evaluation of the Convolutional Neural Networks (CNNs) LeNet-5, AlexNet, and GoogLeNet is made for such task. Here, new validation techniques and a variety of CNNs' hyper-parameters values are considered. We also assess several pre-processing strategies in order to evaluate these CNNs. Moreover, our work presents an analysis of optimization of training hyper-parameters, which can affect the convergence of cost function, the learning speed and the classification performance. Our best results were achieved by GoogLeNet architecture trained with images with contrast stretching and average subtraction resulting in 95.53% of accuracy, with initial learning rate in 0.001 and gamma factor in 0.5.