Towards Generalizable and Robust Virtual Staining via Causal Intervention

Staining pathological tissue is essential for accurate diagnosis but remains a labor-intensive and time-consuming step in pathology workflows. Virtual staining, which computationally generates images resembling conventionally stained tissue sections, offers a promising alternative by significantly improving efficiency and reducing costs. While recent virtual staining methods have shown impressive performance, achieving strong generalization across datasets and maintaining robustness under heterogeneous staining and acquisition conditions remain major challenges. To tackle this issue, we propose a causal intervention-based training framework to improve the generalization and robustness of virtual staining models. From a causal perspective, pathology images contain pathology-relevant features and pathology-irrelevant nuisance variations. While pathology-irrelevant features vary in form and origin, we present a unified strategy to capture nuisance-related variations without explicitly specifying their particular manifestations. Causal intervention is then incorporated during training to suppress these nuisance-related features, encouraging the model to focus on more stable pathology-relevant signals. To the best of our knowledge, this is among the first attempts to enhance the generalization and robustness of virtual staining from a causal perspective. Furthermore, our framework is modelagnostic and can be applied to existing virtual staining models in a plug-and-play manner. Extensive experiments on multiple datasets, staining tasks, evaluation protocols, and experimental settings demonstrate the effectiveness of the proposed method. Code will be available at https://github.com/999Q999L999/Causal-Virtual-Staining.

Authors

Institutions

Publication Details

Journal
IEEE Transactions on Medical Imaging
Published
2026-11-17
DOI
https://doi.org/10.1109/tmi.2026.3755115
Primary Topic
AI in cancer detection
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Towards Generalizable and Robust Virtual Staining via Causal Intervention

Yihuang Hu, Liansheng Wang, Runchen Zhu, Weiping Lin et al.
1 links
IEEE Transactions on Medical Imaging
AI in cancer detection
article

Towards Generalizable and Robust Virtual Staining via Causal Intervention

Yihuang Hu, Liansheng Wang, Runchen Zhu, Weiping Lin, Baoshun Wang, Yiyang Guo
article en

Abstract

Staining pathological tissue is essential for accurate diagnosis but remains a labor-intensive and time-consuming step in pathology workflows. Virtual staining, which computationally generates images resembling conventionally stained tissue sections, offers a promising alternative by significantly improving efficiency and reducing costs. While recent virtual staining methods have shown impressive performance, achieving strong generalization across datasets and maintaining robustness under heterogeneous staining and acquisition conditions remain major challenges. To tackle this issue, we propose a causal intervention-based training framework to improve the generalization and robustness of virtual staining models. From a causal perspective, pathology images contain pathology-relevant features and pathology-irrelevant nuisance variations. While pathology-irrelevant features vary in form and origin, we present a unified strategy to capture nuisance-related variations without explicitly specifying their particular manifestations. Causal intervention is then incorporated during training to suppress these nuisance-related features, encouraging the model to focus on more stable pathology-relevant signals. To the best of our knowledge, this is among the first attempts to enhance the generalization and robustness of virtual staining from a causal perspective. Furthermore, our framework is modelagnostic and can be applied to existing virtual staining models in a plug-and-play manner. Extensive experiments on multiple datasets, staining tasks, evaluation protocols, and experimental settings demonstrate the effectiveness of the proposed method. Code will be available at https://github.com/999Q999L999/Causal-Virtual-Staining.

IEEE Transactions on Medical ImagingVol. PP
Xiamen University (CN)
National Natural Science Foundation of China
Good health and well-being
AI in cancer detection
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon