Efficient 3D Transformer with Cluster-Based Domain-Adversarial Learning for Volumetric Medical Image Segmentation

Real-world application of volumetric medical image segmentation is still challenging due to the domain shift problem and the disability to interpret volumetric information efficiently by current algorithms. To address these problems, we propose a 3D Swin Transformer with a pyramidal downsampling strategy to interpret volumetric information efficiently. Specifically, the novel Transformer includes a spatial downsampling strategy, which downsamples 2D slices pyramidally according to the spatial relationship. Furthermore, we propose a cluster-based domain-adversarial learning algorithm with a cluster-based domain generation algorithm to attenuate the domain shift problem. The cluster-based domain generation algorithm generates numerous cluster-based domains from center-based domains, ameliorating the domain-adversarial learning performance. We evaluate our model against other competitive models on brain stroke lesion segmentation and prostate segmentation tasks. Results indicate that our proposed model outperforms other models on each segmentation task, demonstrating the efficacy of our proposed pyramidally downsampled Transformer block and cluster-based domain-adversarial learning.