STRUDEL: Self-training with Uncertainty Dependent Label Refinement Across Domains

We propose an unsupervised domain adaptation (UDA) approach for white matter hyperintensity (WMH) segmentation, which uses Self-TRaining with Uncertainty DEpendent Label refinement (STRUDEL). Self-training has recently been introduced as a highly effective method UDA, is based on self-generated pseudo labels. However, labels can be very noisy and therefore deteriorate model performance. to predict the uncertainty of integrate it in training process uncertainty-guided loss function highlight high certainty. STRUDEL further improved by incorporating segmentation output existing label generation that showed robustness WMH segmentation. In our experiments, we evaluate standard U-Net modified network higher receptive field. Our results across datasets demonstrate significant improvement respect self-training.