VR-based approach for MCI assessment system using fNIRS and graph convolutional network

Mild Cognitive Impairment (MCI) assessment plays a vital role in identifying cognitive decline, and early intervention can be provided to reduce the risk of dementia. Virtual reality (VR)-based methods have shown promise in MCI assessment due to enhanced engagement, ecological validity, and user-friendliness. Nevertheless, most existing methods focus on MCI-induced behavioural impairment, ignoring the underlying changes in the brain's neural activity. To fill this research gap, we propose a novel approach combining VR and functional near-infrared spectroscopy (fNIRS) for MCI assessment. First, we conducted an experiment involving 21 healthy controls and 12 MCI who participated in two VR tasks while their neural activity was recorded using functional near-infrared spectroscopy (fNIRS). Second, a novel fNIRS-based graph representation was constructed for each subject, incorporating temporal, frequency, and spatial features, where the temporal and frequency features served as node attributes and spatial features as edges. Third, a Graph Convolutional Network (GCN) was employed to enable structure-aware integration of the multidimensional fNIRS graph representations, facilitating the modelling of region-level interactions and enhancing the identification of MCI-related neural alterations. The results showed that the proposed method achieved an MCI classification accuracy of approximately 0.92. The proposed fNIRS-based method combines artificial intelligence (AI) and VR for cognitive impairment screening, with the potential for dementia prevention and the development of intelligent cognitive assessments.