Enhancing cervical squamous cell carcinoma diagnosis under distribution shift via domain-adaptive representation learning
Abstract
Automated whole-slide image (WSI) analysis for cervical squamous cell carcinoma (CSCC) is challenged by limited pathological specificity in general models and poor cross-center generalization. We propose CSCC-PAG (pre-training, adaptation, generalization), a framework integrating domain-specific pre-training with self-supervised domain adaptation. Utilizing 8,041 CSCC WSIs, we applied the self-supervised pre-training for CSCC (SPC) feature extractor, which significantly outperforms pan-cancer baselines in the typing, grading, and segmentation tasks. To address domain shift, we developed representative sample domain adaptation (RSDA), combining masked autoencoders (MAEs) with a k-means buffer strategy for label-free adaptation. The results demonstrate that SPC captures fine-grained CSCC morphology, while RSDA effectively recovers performance degradation across centers without requiring target-domain annotations. This study provides a comprehensive solution for automated CSCC diagnosis, covering the entire pipeline from large-scale pre-training to few-shot generalized deployment, demonstrating its immense potential for real-world clinical implementation.
Related
- Multimodal Detection of a Minute Uterine Cornual Lesion in Cervical Clear Cell Carcinoma
- Beyond HPV16/18: genotype distribution of high-risk HPV and cycle threshold values in cervical carcinogenesis spectrum
- Treatment strategies for unilateral papillary thyroid carcinoma with ipsilateral cervical lymph node metastasis: a narrative review
- Enhancing self-supervised representation learning through lightweight learnable data augmentation
- Synchronous medullary and papillary thyroid carcinoma with mixed-component lateral cervical lymph node metastases: a case report
BibTeX
@article{huang2026enhancing,
title = {{Enhancing cervical squamous cell carcinoma diagnosis under distribution shift via domain-adaptive representation learning}},
author = {Zirui Huang and Ye Liu and Saihui Jin and Yizhi Wang and Mingxi Fu and Jingli Ouyang and Deshui Yu and Taojie Zhu and Fanyi Zeng and Xinyi Pan and Jiatong Ye and Juanjuan Wang and Xiaobo He and Xiao Cheng and Ming Zhao and Yonghong He and Hongfeng Zhang and Bin Hu},
journal = {iScience},
year = {2026},
month = oct,
doi = {10.1016/j.isci.2026.117744},
url = {https://www.cell.com/iscience/fulltext/S2589-0042(26)03125-1},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - Enhancing cervical squamous cell carcinoma diagnosis under distribution shift via domain-adaptive representation learning
AU - Zirui Huang
AU - Ye Liu
AU - Saihui Jin
AU - Yizhi Wang
AU - Mingxi Fu
AU - Jingli Ouyang
AU - Deshui Yu
AU - Taojie Zhu
AU - Fanyi Zeng
AU - Xinyi Pan
AU - Jiatong Ye
AU - Juanjuan Wang
AU - Xiaobo He
AU - Xiao Cheng
AU - Ming Zhao
AU - Yonghong He
AU - Hongfeng Zhang
AU - Bin Hu
PY - 2026
DA - 2026/10/16
JO - iScience
DO - 10.1016/j.isci.2026.117744
UR - https://www.cell.com/iscience/fulltext/S2589-0042(26)03125-1
AB - Automated whole-slide image (WSI) analysis for cervical squamous cell carcinoma (CSCC) is challenged by limited pathological specificity in general models and poor cross-center generalization. We propose CSCC-PAG (pre-training, adaptation, generalization), a framework integrating domain-specific pre-training with self-supervised domain adaptation. Utilizing 8,041 CSCC WSIs, we applied the self-supervised pre-training for CSCC (SPC) feature extractor, which significantly outperforms pan-cancer baselines in the typing, grading, and segmentation tasks. To address domain shift, we developed representative sample domain adaptation (RSDA), combining masked autoencoders (MAEs) with a k-means buffer strategy for label-free adaptation. The results demonstrate that SPC captures fine-grained CSCC morphology, while RSDA effectively recovers performance degradation across centers without requiring target-domain annotations. This study provides a comprehensive solution for automated CSCC diagnosis, covering the entire pipeline from large-scale pre-training to few-shot generalized deployment, demonstrating its immense potential for real-world clinical implementation.
SN - 2589-0042
ER -