An Automated Method with Feature Pyramid Encoder and Dual-Path Decoder for Nuclei Segmentation

Nuclei instance segmentation is a critical part of digital pathology analysis for cancer diagnosis and treatments. Deep learning-based methods gradually replace threshold-based ones. However, automated techniques are still challenged by the morphological diversity of nuclei among organs. Meanwhile, the clustered state of nuclei affects the accuracy of instance segmentation in the form of over-segmentation or under-segmentation. To address these issues, we propose a novel network consists of a multi-scale encoder and a dual-path decoder. Features with different dimensions generated from the encoder are transferred to the decoder through skip connections. The decoder is separated into two subtasks to introduce boundary information. While an aggregation module of contour and nuclei is attached in each decoder for encouraging the model to learn the relationship between them. Furthermore, this avoids the splitting effect of independent training. Experiments on the 2018 MICCAI challenge of Multi-Organ Nuclei Segmentation dataset demonstrate that our proposed method achieves state-of-the-art performance.