en.Wedoany.com Reported - Over 20 hospitals have jointly launched the cloud-based version of the Ruijin RuiPath pathology foundation model, marking a new phase in the large-scale application of medical artificial intelligence. At the "Digital Intelligence Symbiosis, Cloud-Based Launch" Medical AI Innovation Forum, institutions including Ruijin Hospital, the Affiliated Hospital of Yan'an University, Handan Central Hospital, the Affiliated Hospital of Hebei University of Engineering, Shexian County Hospital, Wu'an First People's Hospital, Xingyi People's Hospital, and Ruian People's Hospital participated in the cloud-based launch ceremony for the model.
To address structural challenges in pathological diagnosis, such as the scarcity of high-quality resources, a shortage of pathologists, and regional development imbalances, Ruijin Hospital collaborated with Huawei to develop the RuiPath pathology foundation model. This model is the first clinical-grade pathology foundation model in China to be integrated into hospital production workflows, covering 90% of common cancer types and 90% of downstream diagnostic tasks. Currently, RuiPath follows a deployment pathway of "development and validation at top-tier hospitals—expansion to municipal tertiary hospitals—replication and application at county-level hospitals," continuously adapting and iterating in real clinical scenarios.
Since 2020, Ruijin Hospital has advanced its pathology digitalization efforts, now scanning approximately 5,000 to 6,000 slides daily, with millions of digital pathology slides accumulated. Building on this foundation, Ruijin Hospital partnered with Huawei to progressively construct a RuiPath production workflow covering common cancer types, from lossless compression and unified formats to end-cloud collaborative computing. The model outperforms international counterparts in 7 out of 14 mainstream tasks across 12 open-source datasets. Through a data flywheel mechanism, the model continuously iterates during clinical use, achieving an engineering closed-loop where "accuracy improves with use."
To address engineering challenges in transitioning from laboratory to clinical deployment, Ruijin Hospital and Huawei Cloud have introduced a joint solution featuring an end-cloud collaborative hybrid computing architecture: at the hospital end, features from a small number of suspected tumor regions are extracted and uploaded to the cloud, reducing bandwidth usage by 85% with encrypted transmission; on the cloud, diagnostic reasoning and model retraining are performed under a fully confidential "usable but invisible" mechanism, allowing hospitals to build customized models with only small sample data. This approach alleviates the shortage of computing power and bandwidth at grassroots hospitals while ensuring data and model security, achieving the goal of keeping data within the hospital.
Dr. Tian Yunxiao, Director of the Pathology Department at Handan Central Hospital, shared practical insights, stating that the hospital has completed localized optimization of models for breast cancer and colorectal cancer through small-sample training with just dozens of slides per round: for the breast cancer model, accuracy in distinguishing tumor presence in local Handan cases improved from approximately 95% in early stages to nearly 100%, with histological type accuracy exceeding 90%. For the colorectal biopsy small specimen model, after several rounds of training, accuracy in detecting tumor presence reached 99%, and histological type accuracy exceeded 93%. Leveraging Huawei Cloud's small-sample training and data flywheel mechanism, the hospital required only about 10% of training data to build its customized model.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
