We propose a scalable Temporal Attention Module (TAM) to inject cardiac motion information into segmentation models. TAM captures dynamic changes across temporal frames using multi-headed, cross-temporal attention. It is adaptable across imaging modalities, scalable from 2D to 3D, and adds minimal computational overhead.
mri segmentation deeplearning cardiac-segmentation echocardiography cardiac-motion temporal-attention-module motion-enhanced-segmentation cardiac-image-computing
-
Updated
Jun 24, 2025 - Jupyter Notebook