TriAxial Low-Rank Transformer for Efficient Medical Image Segmentation
A tri-axially factorized low-rank transformer that reduces computational cost while maintaining high segmentation accuracy. (Corresponding author: Xi Fang)
Efficient 3D transformer development and self-supervised pretraining for transferable medical image foundation models.
- Developed an efficient 3D transformer backbone for medical image segmentation with stronger efficiency and accuracy than 3D ViT and Swin Transformer baselines.
- Worked on self-supervised training for 3D transformers to support transfer across classification, reconstruction, registration, and segmentation tasks.