Siamese few-shot network: a novel and efficient network for medical image segmentation

Guangli Xiao, Shengwei Tian, Long Yu, Zhicheng Zhou, Xuanli Zeng

Abstract: Few-shot learning is attracting more researchers due to its outstanding ability to find unseen classes with less data. Meanwhile, we noticed that medical data is difficult to collect and label, but there is a major need for higher accuracy in either organ segmentation or disease classification. Therefore, we propose a few-shot learning model with a Siamese core, the Siamese few-shot network (SFN) to improve medical image segmentation. To the beset of our knowledge, SFN is the first model to introduce few-shot learning combined with the Siamese idea to medical image segmentation. Furthermore, we also design a grid attention(GA) module to locally focus semantic information, especially in medical images. The results prove that our method outperforms the state-of-the-art model on abdominal organ segmentation for CT and MRI.