Uncertainty-guided dual-views for semi-supervised volumetric medical image segmentation

Deep learning has led to tremendous progress in the field of medical artificial intelligence. However, training deep-learning models usually require large amounts of annotated data. Annotating large-scale datasets is prone to human biases and is often very laborious, especially for dense prediction tasks such as image segmentation. Inspired by semi-supervised algorithms that use both labelled and unlabelled data for training, we propose a dual-view framework based on adversarial learning for segmenting volumetric images. In doing so, we use critic networks to allow each view to learn from high-confidence predictions of the other view by measuring a notion of uncertainty. Furthermore, to jointly learn the dual-views and the critics, we formulate the learning problem as a min–max problem. We analyse and contrast our proposed method against state-of-the-art baselines, both qualitatively and quantitatively, on four public datasets with multiple modalities (for example, computerized topography and magnetic resonance imaging) and demonstrate that the proposed semi-supervised method substantially outperforms the competing baselines while achieving competitive performance compared to fully supervised counterparts. Our empirical results suggest that an uncertainty-guided co-training framework can make two neural networks robust to data artefacts and have the ability to generate plausible segmentation masks that can be helpful for semi-automated segmentation processes.