A HEp-2 Cell Image Classification Model Based on Deep Residual Shrinkage Network Combined with Dilated Convolution

Overview & Synopsis

Detection of antinuclear antibodies (ANA) in human epithelial cells (HEp-2) is a common method for the diagnosis of autoimmune diseases. The recognition of fluorescence images of human epithelial cells obtained using Indirect Immunofluorescence (IIF) is a key step in the classification of ANA. To address the problems of low efficiency and high labor intensity caused by manual evaluation methods, a HEp-2 cell fluorescent image classification model based on the depth residual shrinkage network combined with dilated convolution is proposed. First, The model accomplish feature extraction by expanding the field of sensation through dilated convolution. Secondly, a 50-layer deep residual network is build, and each residual module embeds a soft threshold learning sub-network to shrink the output data of the original residual module by the automatically learned soft threshold to achieve the purpose of noise removal. Finally, the multi-scale feature fusion module and fusion of features at different scales are used to complete the classification of HEp-2 cell. Experiments show that this model has good performance and is superior to other depth neural network methods.