AgileFormer: Spatially agile and scalable transformer for medical image segmentation

Peijie Qiu, Jin Yang, Sayantan Kumar, Soumyendu Sekhar Ghosh, Aristeidis Sotiras

Abstract: Highlights•AgileFormer captures spatially varying features in medical image segmentation.•Patch embedding and positional encoding are as crucial as self-attention in ViT-UNet.•AgileFormer achieves SOTA on multi-organ, cardiac, and brain tumor segmentation.•AgileFormer scales well, enhancing segmentation accuracy as model size increases.

External IDs:doi:10.1016/j.bspc.2025.108842