Dec, 2020

自监督学习放射学图像像素级解剖嵌入

TL;DRSelf-supervised Anatomical eMbedding (SAM) is introduced to extract semantic embeddings for imaging pixels with intrinsic structures, which can be used to locate body parts in other images by nearest neighbor searching, surpassing supervised methods trained on 50 labeled images with only one labeled template image and being effective in improving image registration and initializing CNN weights.