Evaluation of tumor budding with virtual panCK stains generated by novel multi-model CNN framework

Xingzhong Hou, Zhen Guan, Xianwei Zhang, Xiao Hu, Shuangmei Zou, Chunzi Liang, Lulin Shi, Kaitai Zhang, Haihang You

Abstract: Highlights•Interpretable Multi-Model FrameworkOur Multi-Model framework can be separate into cytokeratin distribution reconstruction and virtual panCK stain. Each part’s result can be checked separately. Thus it is explainable by the interpretability analysis. This feature is rare in virtual-stain models.•Accurate and ReliableExperiments demonstrate that our virtual panCK could accurately capture subtle structures such as occult tumor cells and tumor budding with high accuracy, which traditional virtual stain methods can not achieve.•Effective and EfficientOur framework is capable producing high quality virtual panCK stains from chemical H&E stains within a short period of time. This feature could reduce the turnaround time of diagnosis from two days to a few hours.•User FriendlyComparing with other digital pathology software, our framework is more friendly to pathologists with little computational background. Together with interpretability and reliability, it could support pathologists for decision-making.

External IDs:dblp:journals/cmpb/HouGZHZLSZY24