HEp-2 cells staining patterns classification via wavelet scattering network and random forest

Automatic classification of Human Epithelial Type-2 (HEp-2) cells staining patterns is a challenging and important problem in the field of medical image analysis. This paper proposed an efficient framework to address the HEp-2 cell image classification problem based on wavelet scattering network and Random Forest. The wavelet scattering network computes rotation-invariant wavelet coefficients as representations of cells images which are robust to different orientations and scale changes in HEp-2 cells images. A Random Forest classifier is trained and then subsequently used to predict the pattern label of a cell image from six classes. Our method is extensively evaluated on two benchmarking datasets based on different experimental protocols. Results show that the proposed system is highly effective on both ICPR 2012 dataset and ICIP 2013 dataset and outperforms several state-of-the-art methods including LBPs, RICLBP, PRICoLBP, and high-order statistic of microtexton.