STRUDEL: Self-Training with Uncertainty Dependent Label Refinement across Domains
This work addresses domain adaptation challenges in medical imaging for WMH segmentation, offering an incremental improvement by reducing label noise in self-training.
The paper tackles the problem of noisy pseudo labels in unsupervised domain adaptation for white matter hyperintensity segmentation by proposing STRUDEL, which uses uncertainty prediction to refine labels and incorporates robust segmentation outputs, resulting in significant improvement over standard self-training methods.
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 for UDA, which is based on self-generated pseudo labels. However, pseudo labels can be very noisy and therefore deteriorate model performance. We propose to predict the uncertainty of pseudo labels and integrate it in the training process with an uncertainty-guided loss function to highlight labels with high certainty. STRUDEL is further improved by incorporating the segmentation output of an existing method in the pseudo label generation that showed high robustness for WMH segmentation. In our experiments, we evaluate STRUDEL with a standard U-Net and a modified network with a higher receptive field. Our results on WMH segmentation across datasets demonstrate the significant improvement of STRUDEL with respect to standard self-training.