Human Brain Connectome Mapping Using Graph Neural Networks Across Applications: A Systematic Review and Diagnostic-Accuracy Sub-study

Abstract

This systematic review will examine the applications of graph neural networks (GNNs) in human brain connectome mapping, including the analysis of structural and functional brain connectivity derived from neuroimaging. It will synthesize the methodological characteristics of eligible studies, neuroimaging modalities, brain atlases, graph construction strategies, GNN architectures, validation procedures, reported performance measures, clinical applications, interpretability, reproducibility, and limitations. The review will follow PRISMA 2020 and PRISMA-S reporting guidance. A diagnostic-accuracy sub-study will be conducted where applicable, following PRISMA-DTA guidance. Eligible diagnostic-accuracy studies selected according to the pre-specified sampling procedure will undergo QUADAS-2 assessment, with diagnostic test accuracy meta-analysis undertaken only when the available evidence is sufficiently comparable. The protocol was prepared before screening and subsequently amended. The amendment history, actual screening status at registration, and any deviations from the protocol will be documented transparently.