Learning Erosional Probability Maps for Nuclei Instance Segmentation

Nuclei instance segmentation is an essential but challenging topic in computer vision and medical image analysis due to the ambiguity of nuclei boundaries, the severe border adhesions, and the diversity of nuclei shapes and sizes. In this paper, we propose a simple yet effective proposal-free framework for accurate nuclei instance segmentation. Specifically, we adopt a deep regression model to learn the mapping from the input images to the nuclei erosional inside probability maps and obtain the final instance segmentation results through straightforward dilation post-processing. We evaluate our framework on the open available 2018 Data Science Bowl nuclei instance segmentation dataset and outperform the state-of-the-art methods.