功能磁共振成像
神经影像学
计算机科学
循环神经网络
功能连接
代表(政治)
人工智能
磁共振成像
神经科学
图形
人工神经网络
医学
生物
理论计算机科学
放射科
政治学
法学
政治
作者
Qing Li,Haixing Dai,Jinglei Lv,Lin Zhao,Zhengliang Liu,Zihao Wu,Xia Wu,Claire D. Coles,Xiaoping Hu,Tianming Liu,Dajiang Zhu
标识
DOI:10.1007/978-3-031-46671-7_6
摘要
Prenatal alcohol exposure (PAE) has garnered increasing attention due to its detrimental effects on both neonates and expectant mothers. Recent research indicates that spatio-temporal functional brain networks (FBNs), derived from functional magnetic resonance imaging (fMRI), have the potential to reveal changes in PAE and Non-dysmorphic PAE (Non-Dys PAE) groups compared with healthy controls. However, current deep learning approaches for decomposing the FBNs are still limited to hand-crafted neural network architectures, which may not lead to optimal performance in identifying FBNs that better reveal differences between PAE and healthy controls. In this paper, we utilize a novel graph representation-based neural architecture search (GR-NAS) model to optimize the inner cell architecture of recurrent neural network (RNN) for decomposing the spatio-temporal FBNs and identifying the neuroimaging biomarkers of subtypes of PAE. Our optimized RNN cells with the GR-NAS model revealed that the functional activation decreased from healthy controls to Non-Dys PAE then to PAE groups. Our model provides a novel computational tool for the diagnosis of PAE, and uncovers the brain’s functional mechanism in PAE.
科研通智能强力驱动
Strongly Powered by AbleSci AI