A semi-supervised segmentation network based on noisy student learning for MICCAI FLARE22 Challenge

Nuo Tong, Kunru Wang, Rui Wu, Huiying Yang, Hu Huang

Keywords: multi-organ segmentation, semi-supervised learning, noisy-student, pseudo-labels

Abstract: Abdominal organ segmentation is very important for clinical applications. However, manually annotating organs from CT scans is time-consuming and labor-intensive. Therefore, it is hard to get access to a large amount of annotated data. Semi-supervised learning is an effective method to use unlabeled data to reduce data labeling, which has become a research hotspot. In this work, we adopt the noisy-student learning method, firstly train the teacher model on the manually labeled data and generate pseudo-labels for the unlabeled data through the model, and then train the student model on both of the manual and pseudo-labeled data, continuously iteratively update to produce the final result. Since No new U-Net (nnU-Net) is the state-of-the-art medical image segmentation method and designs task-specific pipelines for different tasks, we adopt 3D nnUNet as the segmentation model during the experiments.