Our work proposes the Curriculum-based Augmented Fourier Domain Adaptation (Curri-AFDA) and proves to achieve superior adaptation, generalization, and robustness performance for medical image segmentation.
**Motivation**
Medical image segmentation is key to improving computer-assisted diagnosis and intervention autonomy. However, due to domain gaps between different medical sites, deep learning-based segmentation models frequently encounter performance degradation when deployed in a novel domain. Moreover, model robustness is also highly expected to mitigate the effects of data corruption.
**Methodology**
Considering all these demanding yet practical needs to automate medical applications and benefit healthcare, we propose the Curriculum-based Fourier Domain Adaptation (Curri-AFDA) for medical image segmentation. Specifically, we design a novel curriculum strategy to progressively transfer amplitude information in the Fourier space from the target domain to the source domain to mitigate domain gaps and incorporate the chained augmentation mixing to further improve the generalization and robustness ability.
**Performance**
Extensive experiments on two segmentation tasks with cross-domain datasets show the consistent superiority of our method regarding adaptation and generalization on multiple testing domains and robustness against synthetic corrupted data. Besides, our approach is independent of image modalities because its efficacy does not rely on modality-specific characteristics. In addition, we demonstrate the benefit of our method for image classification besides segmentation in the ablation study. Therefore, our method can potentially be applied in various medical applications and yield improved performance.
Paper is available at https://lnkd.in/guYe5SBA
Code is released at https://lnkd.in/gfFAfSQX
Thanks to all collaborators, including An Wang, Mengya Xu, and Prof. Hongliang Ren from CUHK, and Dr. Mobarakol Islam from WEISS, UCL.
