Gland Instance Segmentation by Full Resolution Multi-Scale Dilation Residual Networks

Abstract

Accurate segmentation of gland instances in histology images is essential for pathologists to perform quantitative analysis of the malignancy degree in adenocarcinoma and proceed with subsequent diagnosis. However, gland instances are often close to one another with boundaries that are difficult to discern. Most existing networks lack the ability to distinguish boundaries contextual information, resulting in inaccurate segmentation of adjacent instances. In this paper, we introduce the full resolution multi-scale dilation residual network (FRMDR-Net), which aims to improve the accuracy of gland instances segmentation. The proposed network makes use of multi-scale feature fusion and residual connections to enhance performance. Additionally, asymmetric convolutions are employed in the modules to reduce parameters, enabling better extraction of multi-scale information and improving network efficiency. The experimental results on the 2015 MICCAI Gland Segmentation Challenge dataset show that our proposed method achieves the state-of-the-art performance.