A Deep Residual Inception Network for HEp-2 Cell Classification

Abstract

Indirect-immunofluorescence (IIF) of Human Epithelial-2 (HEp-2) cells is a commonly-used method for the diagnosis of autoimmune diseases. Traditional approach relies on specialists to observe HEp-2 slides via the fluorescence microscope, which suffers from a number of shortcomings like being subjective and labor intensive. In this paper, we proposed a hybrid deep learning network combining the latest high-performance network architectures, i.e. ResNet and Inception, to automatically classify HEp-2 cell images. The proposed Deep Residual Inception (DRI) net replaces the plain convolutional layers in Inception with residual modules for better network optimization and fuses the features extracted from shallow, medium and deep layers for performance improvement. The proposed model is evaluated on publicly available I3A (Indirect Immunofluorescence Image Analysis) dataset. The experiment results demonstrate that our proposed DRI remarkably outperforms the benchmarking approaches.