项目组前期功能磁共振研究初步验证了恐惧记忆环路是创伤后应激障碍 (PTSD) 发病的敏感和特异脑神经网络，但传统脑影像方法并不能作为生物学标记物实现个体水平PTSD的准确诊断，关于恐惧记忆神经网络涉及脑区及机制也尚未有明确结论。本项目拟采用结合机器学习的多模态磁共振脑网络分析技术，构建以恐惧记忆环路为关键节点的结构连接网络(3D结构像)、白质纤维脑网络(弥散张量成像)和功能连接网络(静息态功能磁共振)，并进行结构-功能耦合分析；采用稀疏编码的特征提取及多核线性支持向量机等机器学习算法，提取影像特征指标、建立PTSD早期诊断模型；采用结合LASSO和随机森林的回归模型进行多模态特征和PTSD认知评分之间回归分析，寻找特异性预测PTSD认知评分影像学标志物，探索PTSD 认知损伤同功能评价之间的关系机制；有利于阐明恐惧记忆脑网络对PTSD的调控机制，促进高级MRI 脑影像技术的临床转化。

Our previous functional MRI research preliminarily manifestoed that the fear memory network is sensitive and specific for post-traumatic stress disorder (PTSD), but the traditional neuroimaging method can’t be used as a biomarker to precise diagnosis on an individual level. The whole regions and the mechanisms of fear memory network were still unclear until now. This project proposes a multi-modal MRI brain network analysis combined with machine learning strategy. Based on the key nodes of fear memory network, morphometric MRI is used to construct structural connectivity network (SCN), diffusion tensor imaging is used to construct the anatomical white matter fiber network, the resting-state functional MRI is used to construct functional connectivity network (FCN), the structure-function coupling network also is analyzed. Furthermore, propose a combination model of sparse coding (SC) and multi-kernel linear support vector machine (MKL-SVM) to identify the discriminative features and perform the early diagnosis of PTSD; propose a regression model by combination of the least absolute shrinkage and selection operator (LASSO) and proximity-based random forest (RF) for predicting clinical scores and identifying the most correlated biomarkers of PTSD. This project will shade lights on the fear memory network on the regulation of PTSD and will promote clinical translation of advanced MRI techniques.