.png){kind=icon}
.png){kind=icon}
Return
Brain Age Estimation Based on Age-Specific Graph Structure Learning
.png){kind=icon}
PRE
AI
DOI：10.1117/12.3095235.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Utilizing structural magnetic resonance imaging (sMRI) to capture brain structural changes for brain age prediction is beneficial for research on brain aging, as well as early screening and intervention of brain-related diseases. Given the complex topological relationships between brain regions, existing studies typically employ graph neural networks for brain age prediction. However, aging mechanisms, involved brain regions, and aging progression features differ across age groups, and existing methods fail to extract aging features in an age-specific manner. To solve these problems, this paper proposes an Age-specific Graph Structure Learning network (Age-Specific-GSL), which dynamically constructs optimal brain network structures guided by age-specific universal structural to enhance prediction accuracy. First, a multi-level subgraph division module is designed to focus on key subnetworks by leveraging persistent aging features and age-specific differential features, reducing the complexity of learning universal structures. Second, an age-specific feature-guided graph structure learning network is developed to minimize the impact of individual variability and construct individual brain network structures strongly correlated with aging. Finally, by comprehensively analyzing the optimal graph structures and brain region features, accurate brain age estimation is achieved. Experimental results show that Age-SpecificGSL achieves an average mean absolute error (MAE) of 5.83 years across two public sMRI datasets, reaching state-of-the-art performance. Moreover, the graph structures incorporating age-specific aging features better align with the dynamic patterns of brain aging. Based on this, we further analyze differences in inter-regional connectivity strength across age groups, providing deeper insights into the brain aging process.
Keywords:
Brain age estimation
Graph neural network
sMRI
Brain network
Journal
S
IF:
0
Papers:
75
Citations:
0