Advancing Functional Near-Infrared Spectroscopy (fNIRS) With Transformers: A PRISMA-Driven Systematic Review

Functional Near-Infrared Spectroscopy (fNIRS) has emerged as a promising neuroimaging modality in various domains, including cognitive neuroscience, brain-computer interfaces (BCIs), clinical diagnostics, mental workload assessment, and neurorehabilitation. Despite its growing applications, processing fNIRS signals presents unique challenges. The advent of deep learning, particularly Transformers, has taken on an increasingly pivotal role in addressing these complexities. As Transformers have revolutionized multiple fields, their application in fNIRS signal processing has gained significant traction recently. To support ongoing research, this study systematically reviews the advancements and applications of Transformer models in fNIRS research, following PRISMA guidelines. A comprehensive analysis was conducted on 30 studies (N = 895), offering a structured evaluation of current trends, methodologies, and outcomes. This review paper highlights Transformer architectures in fNIRS-based studies, focusing on advanced modifications in key stages such as the embedding and attention mechanisms. Multimodal studies analyzing fNIRS with other modalities, such as Electroencephalography (EEG), using Transformers are also reviewed. Finally, the paper examines open-access and private datasets used in the literature and offers guidelines for future research, identifying open problems that warrant further exploration.