fNIRS in autism neuroscience: From replication to discovery

Studying the neural basis of autism spectrum disorder (ASD) has long been hampered by a mismatch: core social features unfold during interaction, whereas standard neuroimaging commonly relies on isolated, scanner-based environments. Functional near-infrared spectroscopy (fNIRS) partly addresses this mismatch. It is portable, more tolerant of moderate movement than fMRI, and compatible with live social interaction, although it remains sensitive to optode motion and systemic physiology and is restricted primarily to superficial cortex. This narrative review examines three contributions of fNIRS to ASD neuroscience: access to developmental populations that are difficult to study with fMRI; dyadic paradigms that measure neural coordination during live interaction; and integration with behavioral, gaze, autonomic, and electrophysiological signals. We then review wearable and naturalistic systems, machine-learning-based classification, and applications across related neurodevelopmental conditions, while critically appraising small samples, task heterogeneity, and preprocessing variability. Among the 25 reports examined in detail, no classification model had been evaluated in an independent external ASD cohort. Eleven reports linked an fNIRS-derived measure to a clinical, behavioral, or developmental outcome through at least one statistically significant association. These associations were generally exploratory, cross-sectional, and not independently validated. The evidence therefore supports fNIRS as a research tool for cortical hemodynamic measurement, while specific fNIRS-derived patterns of regional activation or connectivity have yet to be validated as biomarkers. Priorities include longitudinal multi-site cohorts, physiologically informed preprocessing, population-specific motion validation, independent model testing, and direct linkage of neural measures to developmental or treatment outcomes.