AG-TAL
AG-TAL improved performance across CNN, Transformer-based, and Mamba-based backbones. AG-TAL is added to cross-entropy loss as a composite objective with radius-aware Dice, breakage-aware clDice, and adjacency-aware co-occurrence component…
1 sources - 4 claims
AG-TAL improved performance across CNN, Transformer-based, and Mamba-based backbones. AG-TAL is added to cross-entropy loss as a composite objective with radius-aware Dice, breakage-aware clDice, and adjacency-aware co-occurrence components. AG-TAL combines anatomical and topological priors to improve computationally efficient multiclass Circle of Willis segmentation. AG-TAL is intended to function without architecture changes or inference pipeline changes.