Differentiate preterm and term infant brains and characterize the corresponding biomarkers via DICCCOL-based multi-modality graph neural networks

• DOI: 10.3389/fnins.2022.951508
• Publication Date: 2022-10-14T05:31:57Z
• Authors: Shu Zhang; Ruoyang Wang; Junxin Wang; Zhibin He; Jinru Wu; Yanqing Kang; Yin Zhang; Huan Gao; Xintao Hu; Tuo Zhang
• Keywords: Pooling

Preterm birth is a worldwide problem that affects infants throughout their lives significantly. Therefore, differentiating brain disorders, and further identifying characterizing the corresponding biomarkers are key issues to investigate effects of preterm birth, which facilitates interventions for neuroprotection improves outcomes prematurity. Until now, many efforts have been made study birth; however, most studies merely focus on either functional or structural perspective. In addition, an effective framework not only jointly function structure at group-level, but also retains individual differences among subjects. this study, novel dense individualized common connectivity-based cortical landmarks (DICCCOL)-based multi-modality graph neural networks (DM-GNN) proposed differentiate term infant brains characterize biomarkers. This adopts DICCCOL system as initialized node GNN each subject, utilizing both profiles effectively retaining differences. To be specific, magnetic resonance imaging (fMRI) provides features nodes, fiber connectivity utilized representation edges. Self-attention pooling (SAGPOOL)-based then applied identify Our results successfully demonstrate can brains. Furthermore, self-attention-based mechanism accurately calculate attention score recognize significant 87.6% classification accuracy observed developing Human Connectome Project (dHCP) dataset, distinguishing explored extracted. uniform disorders