AMN: An Adaptive Multi-Scale Fusion Network with Boundary and Uncertainty Modeling for Nuclei Segmentation
Researchers have developed AMN, an Adaptive Multi-Scale Fusion Network designed for precise nuclei segmentation in histopathology images. This dual-encoder framework uniquely combines a Swin Transformer and a ResNet-50 feature pyramid, using a learned gating mechanism to dynamically balance their contributions across different scales. AMN incorporates a multi-objective loss function that includes focal loss, boundary-aware loss, and an uncertainty-modulated classification term to improve accuracy and reduce overconfident errors. The model achieved state-of-the-art results on the CoNIC benchmark, outperforming eight other architectures and demonstrating robust generalization capabilities on the MoNuSeg dataset. AI
IMPACT This novel segmentation approach could enhance diagnostic accuracy in computational pathology, potentially improving treatment planning and prognosis prediction.