whole-slide-images

BigCell是首个能在原生千兆像素分辨率下生成连贯全切片病理图像的方法，采用流匹配Transformer在2048×2048像素上训练超过12万张公开WSI，可生成高达32,768×32,768像素的图像，并通过合成样本将少样本切片级分类性能提升最多18%。

SlideBank is a training-free framework for whole-slide image reasoning in pathology, using a persistent hierarchical evidence bank to improve consistency and reduce inference costs.

Introduces AdaSurvMamba, an adaptive framework for multimodal survival analysis that uses a dual-scale importance-aware reconstruction module and semantic aggregation scanning to improve integration of whole slide images and genomic profiles, achieving consistent gains across five TCGA cohorts.

This paper proposes TopoMamSurv, a Graph Mamba framework for whole-slide image survival analysis that uses topology-aware ordering to address Mamba's sensitivity to input order, and incorporates bidirectional Mamba and GCN for spatial context modeling.