Semi-Supervised Deep Learning for Fully Convolutional Networks
This addresses the costly annotation issue in image segmentation for medical imaging, but it is incremental as it adapts an existing semi-supervised concept to a new architecture.
The paper tackles the problem of requiring large labeled datasets for deep learning by developing a semi-supervised learning method for Fully Convolutional Networks (FCNs), reporting substantial improvements in MS Lesion Segmentation through domain adaptation.
Deep learning usually requires large amounts of labeled training data, but annotating data is costly and tedious. The framework of semi-supervised learning provides the means to use both labeled data and arbitrary amounts of unlabeled data for training. Recently, semi-supervised deep learning has been intensively studied for standard CNN architectures. However, Fully Convolutional Networks (FCNs) set the state-of-the-art for many image segmentation tasks. To the best of our knowledge, there is no existing semi-supervised learning method for such FCNs yet. We lift the concept of auxiliary manifold embedding for semi-supervised learning to FCNs with the help of Random Feature Embedding. In our experiments on the challenging task of MS Lesion Segmentation, we leverage the proposed framework for the purpose of domain adaptation and report substantial improvements over the baseline model.