Ö. Çiçek, A. Abdulkadir, S. S. Lienkamp, T. Brox, O. Ronneberger. 3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation. MICCAI, 2016.

Cicek et al. present a 3D convolutional network architecture for volumetric segmentation in the biomedical domain from sparse annotations. In particular, they propose the architecture shown in Figure 1, closely related to earlier work [1]. The network combines a decode stage, consisting of several convolutional and pooling layers, with a decoder part consisting of a series of convolutional and up-convolutional layers. Key to their model is that the loss function disregards voxels labeled as "unlabeled". Therefore, the network can be trained using volumes where only a few slices are densely labeled. They further utilize heavy data augmentation including rotations, gray value augmentation and smooth deformations.

Figure 1 (click to enlarge): Illustration of the network architecture.

  • [1] O. Ronneberger, P. Fischer, T. Brox. U-net: Convolutional networks for biomedical image segmentation. MICCAI, 2015.
What is your opinion on this article? Let me know your thoughts on Twitter @davidstutz92 or LinkedIn in/davidstutz92.