A tumor classification method and system based on federated learning and multimodal fusion

By employing a federated learning and multimodal fusion-based tumor classification method, and utilizing collaborative training between a central server and medical nodes, the method addresses the issues of low efficiency and insufficient security in the management of tumor clinical records. It achieves effective integration of multimodal data and privacy protection, thereby improving the accuracy and security of tumor classification.

CN122135787APending Publication Date: 2026-06-02BEIJING YIYONG TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
BEIJING YIYONG TECH CO LTD
Filing Date
2026-03-02
Publication Date
2026-06-02

AI Technical Summary

Technical Problem

Existing technologies rely on manual annotation for the management of tumor clinical records, resulting in low efficiency. Federated learning suffers from high communication overhead, difficulty in model convergence, and insufficient security in cross-institutional collaborations, making it difficult to effectively integrate multimodal tumor data.

Method used

A tumor classification method based on federated learning and multimodal fusion is adopted. The global multimodal tumor model and initialization parameters are sent through the central server. Each medical node is trained and transmits local parameter gradients and feature sets. The central server calculates dynamic weights to update the global model. Combined with feature extraction and privacy protection processing of images, genomes, pathology and medical record texts, the weight allocation and reward value mechanism are optimized by using RL neural network.

Benefits of technology

It improves the accuracy and safety of tumor classification, solves the problem of information silos, adapts to differences in data distribution, improves model convergence speed and training efficiency, and enhances privacy protection performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122135787A_ABST
    Figure CN122135787A_ABST
Patent Text Reader

Abstract

This invention relates to a tumor classification method and system based on federated learning and multimodal fusion, belonging to the field of intelligent medical technology. It solves the problems of data silos, the difficulty in balancing privacy protection and model performance in existing tumor clinical data processing technologies. The method includes: a central server sending a global multimodal tumor model and initial global model parameters to each medical node; each medical node using first local multimodal tumor data to train a local multimodal tumor model, determining local parameter gradients and local feature sets, and sending it to the central server; the central server determining the aggregate parameter gradient, updating the global model parameters, obtaining a trained global multimodal tumor model, and sending it to each medical node; and the medical nodes performing tumor classification based on the trained global multimodal tumor model. This invention achieves joint modeling of tumor data across institutions while ensuring patient privacy and security.
Need to check novelty before this filing date? Find Prior Art