On Direct Distribution Matching for Adapting Segmentation Networks
This addresses the problem of domain shift in medical image segmentation for researchers and practitioners, offering a more stable and efficient alternative to adversarial methods.
The paper tackles domain adaptation for segmentation networks by proposing a direct kernel density matching loss that avoids adversarial training, achieving significantly better accuracy and stability in adapting brain segmentation across MRI modalities.
Minimization of distribution matching losses is a principled approach to domain adaptation in the context of image classification. However, it is largely overlooked in adapting segmentation networks, which is currently dominated by adversarial models. We propose a class of loss functions, which encourage direct kernel density matching in the network-output space, up to some geometric transformations computed from unlabeled inputs. Rather than using an intermediate domain discriminator, our direct approach unifies distribution matching and segmentation in a single loss. Therefore, it simplifies segmentation adaptation by avoiding extra adversarial steps, while improving both the quality, stability and efficiency of training. We juxtapose our approach to state-of-the-art segmentation adaptation via adversarial training in the network-output space. In the challenging task of adapting brain segmentation across different magnetic resonance images (MRI) modalities, our approach achieves significantly better results both in terms of accuracy and stability.