Re-Unet: Multi-Modality Cell Segmentation based on nnU-Net Pipeline

Haotian Lu, Jinghao Feng, Zelin Peng, Wei Shen

Abstract: Cell segmentation is an important initial task in medical image analysis, and in recent years, data-driven deep learning methods have made groundbreaking achievements in this field. In this challenge, a multi-modal and partially labeled dataset is provided. In this paper, we propose a multi-modality cell segmentation framework called Re-Unet, which is based on the nnU-Net pipeline and an iterative self-training method. Re-Unet enriches the original data and fully considers the information of cell intervals while making full use of the semi-supervised data. Our proposed method achieves a mean F1 score of 0.6101 on the tuning set and a F1 score of 0.4492 on the testing set.