Peking University Collaborates with CUHK, Tsinghua University, and CAS on a Federated Learning-Based Cross-Institutional Intelligent Medical

Date: April 7, 2025

Views: [Insert view count]

From March 24 to April 3, 2025, a multi-institutional research collaboration project on federated learning in intelligent healthcare, led by the Institute of Medical Technology at Peking University, achieved significant progress. The project involves four leading institutions: Peking University, The Chinese University of Hong Kong (CUHK), Tsinghua University, and the Institute of Information Engineering, Chinese Academy of Sciences (CAS). Together, the teams have jointly discussed technical challenges in multimodal medical data fusion, heterogeneous data alignment, and privacy-preserving computation.

A cloud-based platform was successfully developed to support the upload and sharing of DICOM-format medical imaging data across multiple sites, achieving PACS-RIS integration via cloud. In addition, a federated learning server was deployed and tested in a multi-center setting, equipped with FPGA acceleration cards to ensure secure communication. A preliminary academic manuscript on AI-assisted diagnosis has been completed as part of the collaboration. The adaptive multimodal federated fusion strategy deployed in this project has significantly improved the accessibility and generalizability of federated learning in real-world medical scenarios. In particular, the pediatric pneumonia risk prediction model tested during the project achieved an accuracy exceeding 80%, clearly outperforming single-center approaches.

This collaboration brought together cross-disciplinary strengths in medical technology, clinical medicine, artificial intelligence, and data security, forming an innovative model of “medical-engineering integration and multi-regional cooperation.” Looking ahead, the project team will continue to refine federated learning techniques and expand their application to areas such as rare disease diagnosis, drug discovery, and telemedicine. The collaboration also aims to contribute to the development of technical standards, providing core technologies for the construction of China’s smart healthcare infrastructure.

Group photos

---

Background

In the field of medical artificial intelligence, data privacy protection and cross-institutional collaboration have long been critical challenges. Traditional centralized machine learning requires the sharing of raw data, which poses significant privacy risks. At the same time, the phenomenon of “data silos” across hospitals hampers the optimization of AI models.

To address these issues, the research team from Peking University, in collaboration with CUHK, Tsinghua University, and CAS Institute of Information Engineering, has adopted a federated learning approach to develop a distributed medical data analysis platform. This technology allows all participating institutions to collaboratively train AI models without sharing raw data, relying instead on encrypted parameter exchange. This preserves patient privacy while maximizing the value of multi-center data resources, offering a replicable model for secure medical data sharing and collaborative AI development.