S^3CNet: Semi-Supervised Spatial-Consistency Constrained Networks for Cardiac MRI Segmentation

Precise cardiac MRI segmentation is crucial for cardiac function analysis.Currently data-driven neural network greatly promotes the development of cardiac MRI segmentation,whereas the dependence on annotated data has limited the generalization of deep neural networks.For the purpose of reducing the dependence on annotated data for deep neural networks,we propose a semi-supervised spatial-consistency networks(S3 CNet)for cardiac MRI segmentation,which employs the spatial transformation consistency of unannotated data before and after being fed into a network architecture.We evaluate the performance of our proposed method on ACDC 2017 cardiac multi-structural segmentation dataset,demonstrating that our method can achieve much better generalization performance than that with supervised learning,meanwhile,the generalization of the proposed method is also better than other state-of-the-art semi-supervised methods.