.png){kind=icon}
.png){kind=icon}
Return
Multiplex aggregation combining sample reweight composite network for pathology image segmentation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.artmed.2025.103239.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• Innovative Network Architecture: Introduced CDNet, a novel architecture combining CNN and Vision Transformer (ViT) to enhance feature edge and texture details, significantly improving nuclei segmentation in pathological images. • Diversified Aggregation Convolution (DAC) Module: Developed the DAC module to integrate feature maps from different downsampling methods, enhancing feature representation and accuracy in handling complex boundaries. • Causal Inference Module (CIM): Integrated CIM based on causal inference principles to eliminate spurious correlations between features, thereby improving the model’s cross-domain generalization and stability. • Stable-Weighted Combined Loss Function: Designed a unique loss function combining Causal Inference Loss, chunk-computed Dice Loss, and Focal Loss to optimize segmentation performance and enhance the model’s robustness. • Superior Segmentation Performance: Demonstrated that CDNet outperforms state-of-the-art models on multiple datasets (MoNuSeg, GLySAC, MoNuSAC), effectively addressing challenges such as blurred boundaries, domain inconsistencies, and uneven nuclei distribution. • Clinical Applicability: Extended CDNet to support both nuclei segmentation and classification, enabling accurate identification and quantification of different nuclear subtypes. This facilitates tissue typing, tumor grading, and the assessment of spatial cell distribution—key factors in clinical diagnosis and personalized treatment planning.
Keywords:
CDNet
Vision Transformer
nuclei segmentation
causal inference
loss function
Journal
IF:
6.2
Papers:
2.5K
Citations:
7.8K
Organization
View more organizations