HEp-2 Cell Classification Using an Ensemble of Convolutional Neural Networks

초록

The presence of Anti-nuclear Autoantibodies (ANA) in human serum is related to autoimmune diseases. Indirect Immunofluorescence (IIF) imaging on human epithelial type-2 cells (HEp-2) is the gold standard for the ANA test. Accurate Human Epithelial-2 (HEp-2) cell classification plays an essential role in the diagnosis of immune system diseases. Developing computer-aided diagnosis (CAD) systems for ANA analysis is necessary for improving the disease diagnosis and treatment planning of the patients. Traditionally, cell patterns are assessed manually with a fluorescence microscope. However, due to the large variations of cell patterns, manual interpretation of the cell images is time-consuming, highly subjective and also requires experienced experts. In this paper, we propose a deep learning-based ensemble model of InceptionV3 and Xception architectures for the task of HEp-2 cell images classification as either healthy or case subjects. In intuition, the aggregation of different architectures can effectively extract and fuse the most discriminative deep features from input images. Also, we use a transfer learning strategy and hyper-parameter tuning to further improve the performance of the proposed model. Experimental results demonstrate that our proposed ensemble model achieves promising results with an accuracy of 95.07%, sensitivity of 99.96% and specificity of 99.79% in comparison with state-of-the-art models on the ICPR 2012 benchmark dataset.

키워드