基于功能连接的大脑年龄预测的影响因素

Influential factors of brain age prediction based on functional connectivity

目的探索基于静息态功能连接进行大脑生理年龄预测的可行性及相关影响因素。方法选取来自阿尔茨海默病神经影像学计划(Alzheimer disease neuroimaging initiative,ADNI)数据库的41例满足条件的健康受试者。首先对静息态功能磁共振成像(resting-state functional MRI,rs-fMRI)数据进行预处理并提取功能连接特征,利用基于Bootstrap的特征筛选方法进行特征降维;然后使用支持向量回归建立正常人大脑年龄的预测模型,最后用留一法进行交叉验证,并比较不同大脑模板、全脑信号回归及性别因素对年龄预测的影响。结果基于AAL-90、AAL-1024、Shen-268、Fan-246脑图谱得到的预测值与真实年龄之间的相关系数r分别为0.23、0.29、0.17、0.38。使用全局信号回归,基于Fan-246脑图谱得到年龄预测模型的相关系数r显著提升为0.66。利用性别分组建模,基于Fan-246脑图谱预测模型的相关系数r提升为0.46。结论根据静息态功能磁共振的功能连接特征可以较好地估计健康大脑的生理年龄,且大脑模板、全局信号回归对年龄估计模型的性能有较大影响。本研究可加深对大脑老化过程的认识,对阿尔兹海默病的早期诊断和预防有着重要的指导价值。

Object To explore the feasibility and related influential factors in brain age prediction based on resting-state functional connectivity.Methods Forty-one eligible healthy subjects were enrolled from the open source Alzheimer disease neuroimaging initiative(ADNI)dataset in this study.After the preprocess of resting-state functional MRI(rs-fMRI)data and the extraction of functional connectivity,the feature selection method based on Bootstrap was used for feature reduction,which was then input into the support vector regression model to construct the age estimation model of healthy brains.The leave-one out cross validation was applied for model evaluation,and the influences of different brain atlases,global signal regression and gender on the estimation performances were also compared.Results The correlational values between the estimated age and the real age were 0.23,0.29,0.17,0.38 respectively for AAL-90,AAL-1024,Shen 268 and Fan-246 atlases.When using the global signal regression,Fan-246 atlas could achieve a significantly improved performance as the correlational value r was increased to 0.66.When using gender information to construct the gender-specific age estimation model,the performance of Fan-246 atlas could increase to 0.46.Conclusions The functional connectivity features from resting-state functional MRI can well estimate the physiological age of healthy brains,and brain atlas and global signal regression are important factors for the age estimation model.This study can deepen the understanding of the aging process in the brain,and provide valuable guidance to the early detection and prevention of Alzheimer disease.