Answers are grounded in this paper and cite their sources.
Electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) are widely used for noninvasive brain monitoring in clinical and cognitive applications, where machine learning-based classification has become increasingly prevalent. Alongside algorithmic advances, a growing body of literature reports signal-derived metrics as digital biomarkers; however, their interpretation, validity, and comparability remain strongly dependent on target definition, processing pipelines, and evaluation practices. This review provides a structured overview of EEG- and fNIRS-based classification studies, with a particular focus on signal processing workflows and biomarker extraction. The reviewed literature is organized around key stages of the analytical pipeline, including target formulation, data preprocessing, feature extraction, and validation. Common feature families-spanning spectral, temporal, connectivity-based, network-level, and hemodynamic representations-are summarized and discussed in relation to their typical applications and limitations. The main contribution is an evidence-based distinction between signal features and credible candidate biomarkers. Reliability, analytical robustness, external generalization, and interpretability are considered within the intended context of use. Clinical utility is assessed separately from predictive performance. Across 117 primary studies, most measures remained candidate biomarkers; only four studies were assigned to independent clinical validation, with qualifications, and none evaluated clinical utility.
No takes yet. Share an insight, caveat, or question.
Kahankova et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: