CerS-Path Cervical Pathology System Published: Pretrained on 140,000 Slides, 99.38% Screening Sensitivity in 3,173 Prospective Cases Across Five Centers
Nature Communications on November 14 published a cervical histopathology study introducing CerS-Path, a Cervical Subspecialty Pathology diagnostic system built on a self-supervised learning framework for cervical cancer pathology.
According to the abstract, CerS-Path was pretrained in a self-supervised manner on about 190 million tissue patches drawn from roughly 140,000 whole-slide images, with multimodal enhancement from 2.5 million image-text pairs. It provides eight diagnostic functions: screening, grading, subtyping, quantification, rare-cancer classification, predictive modeling, multimodal question answering and automated report generation. The authors note that foundation models are promising for cervical cancer screening but face important limits in clinical deployment, and CerS-Path was designed for subspecialty diagnostic work in cervical histopathology.
In prospective validation across five centers covering 3,173 cases, CerS-Path achieved a screening sensitivity of 99.38 percent. According to the abstract, the system surpasses prior foundation models in scope and clinical applicability.
The paper, 'A subspecialty diagnostic system powered by self-supervised learning for cervical histopathology', is authored by Yizhi Wang, Li Chen, Qiang Huang, Tian Guan and colleagues, with participating institutions including Tsinghua Shenzhen International Graduate School and the Ningbo Clinical Pathology Diagnosis Center.
Source: Nature Communications
Published by GSM · USA (original Chinese edition)
Photo: 图：宫颈鳞状细胞癌的组织病理图像（资料图）。来源：Li H 等，Wikimedia Commons，CC BY 4.0。 / Photo: Histopathology image of cervical squamous cell carcinoma (file image). Source: Li H et al., Wikimedia Commons, CC BY 4.0.
