Cooperative Training and Latent Space Data Augmentation for Robust Medical Image Segmentation

Deep learning-based segmentation methods are vulnerable to unforeseen data distribution shifts during deployment, e.g. change of image appearances or contrasts caused by different scanners, unexpected imaging artifacts etc. In this paper, we present a cooperative framework for training models and latent space augmentation method generating hard examples. Both contributions improve model generalization robustness with limited data. The consists fast-thinking network (FTN) slow-thinking (STN). FTN learns decoupled features shape reconstruction tasks. STN priors correction refinement. two networks trained in manner. generates challenging examples masking the both channel-wise spatial-wise manners. We performed extensive experiments on public cardiac datasets. Using only 10 subjects from single site training, demonstrated improved cross-site performance, increased against various compared strong baseline methods. Particularly, yields 15% improvement terms average Dice score when standard method.