基于图强化学习的模态解耦脑龄预测模型

打开文本图片集

中图分类号：TP18 文献标识码：A 文章编号：2096-4706（2026）03-0063-08

Modal Decoupling Brain Age Prediction Model Based on Graph Reinforcement Learning

SONGJiaying，MAHuibin （School ofInformationandElectronic Technology，Jiamusi University，Jiamusi54oo7，China）

Abstract：Inorderto improve the precisionand generalizationabilityof thebrainage prediction model，andto support the early dentifcationofneurodegenerative diseasesandtheresearchonbrainaging mechanisms，this paper proposesamodal decoupling brainage prediction model based ongraphreinforcementlearning.First，this studyconstructs individual brain networksbasedon fMRIand sMRIdata，and uses Graph Neural Network to model the topological features ofbrain regions. Second，the modeladaptivelyadjusts thenumberofgraphconvolutionlayers through thedynamic graphconvolution mechanism （ACframework），andadopts the Double Deep Q-Network （DDQN）tooptimize the GraphSAGE strategy toadapt to brain network paternsof diferent modalities.Experimentalresultsindicate thatthe performanceofthe proposedmodel in metrics suchas MAEissuperiortoexisting depconvolutionalnetworksandtraditional GraphNeural Network methods.This studynot onlyreflects the advantages of dynamic graphconvolutionandreinforcementleaing strategies inbrainage prediction，butalso provides a new technical approach for further exploring the mechanism of brain aging.

Keywords： brain age prediction; Graph Neural Network; reinforcement leaming; graph reinforcement learning; resting state fMRI; structural MRI

0 引言

伴随人类年龄的增长，大脑可能会出现偏离正常发展轨迹的迹象，这类迹象是大脑异常的重要信号。（剩余9389字）