A brain network representation method based on graph convolution and attention mechanism

By constructing a dynamic functional connectivity network using graph convolution and attention mechanisms to represent brain networks, we solve the problem of insufficient utilization of spatiotemporal feature complementarity in existing methods, and achieve more accurate brain disease classification and dynamic characteristic analysis.

CN117593269BActive Publication Date: 2026-06-26ANHUI NORMAL UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
ANHUI NORMAL UNIV
Filing Date
2023-11-22
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

Existing brain network analysis methods fail to effectively utilize the complementary relationship between the temporal and spatial characteristics of brain activity, resulting in information loss and an inability to fully understand the mechanisms of brain diseases.

Method used

By employing graph convolutional networks and attention mechanisms, and constructing a dynamic functional connection network, temporal and spatial features are extracted separately, and feature interaction and fusion are performed to form a global spatiotemporal feature representation.

Benefits of technology

By leveraging the complementary nature of spatiotemporal features through deep learning frameworks, the accuracy of brain disease classification and the ability to understand the dynamic characteristics of brain networks have been improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117593269B_ABST
    Figure CN117593269B_ABST
Patent Text Reader

Abstract

The application provides a brain network representation method based on graph convolution and attention mechanism, time windows are uniformly divided in a time sequence of resting state functional magnetic resonance imaging data, a functional connection matrix is constructed in each time window, and a dynamic functional connection network is constructed. In the dynamic functional connection network, a graph convolution network and an attention mechanism are used to extract time and space features respectively, and twice feature interaction is performed in feature extraction. In this way, from local feature coupling to higher level feature coupling, and finally obtaining global spatio-temporal feature representation, the loss of spatio-temporal feature complementarity caused by the separation of spatio-temporal features in the traditional method can be effectively made up, and the complementary nature of spatio-temporal features can be better utilized.
Need to check novelty before this filing date? Find Prior Art