Neural Noise- EEG and fNIRS are combined in a multimodal pipeline (e.g. EEG-fNIRS-SDS) for neural time-series-based depression recognition.1
- Multimodal EEG–fNIRS methods provide direct relevance for physiological signal classification in clinical and methods applications.1 1
Weekly enrichment (2026-07-20)
- The study proposes MI-WNet, a Mutual Information Maximization-based Weighted Multimodal Fusion Network that jointly analyzes EEG, fNIRS, and Self-Rating Depression Scale (SDS) clinical-scale data for depression recognition.2 3
- MI-WNet reached a depression recognition accuracy of 95.52% ± 0.42%, with recall 95.09% ± 0.68% and specificity 95.80% ± 1.03%.2 3
- Modality-specific backbones feed the fusion network: a TSTPN model for EEG and a Spatio-Temporal Autoencoder Network (STAE-Net) for fNIRS, with STAE-Net reported to outperform traditional machine-learning methods on AUC and specificity.3
- Fusion uses mutual-information maximization plus a dynamic weighting mechanism to filter redundant features and strengthen cross-modal interaction, outperforming single-modal and fixed-weight multimodal baselines.2 3
- Rationale: conventional interview/questionnaire diagnosis is subjective, so the method combines EEG’s high temporal resolution with fNIRS cortical-hemodynamic (metabolic) information and behavioral SDS profiling.2
- Published in the Journal of Neuroscience Methods (2025), DOI 10.1016/j.jneumeth.2025.110668.3
- Context: an independent EEG–fNIRS passive-BCI study using SincShallowNet on 16 EEG + 8 fNIRS channels reported balanced accuracy around 90.9% (auditory condition), corroborating gains from electro-hemodynamic fusion.4
- BCI implication: adaptive multimodal fusion supports objective, data-driven depression screening and passive brain-computer-interface mental-state monitoring rather than subjective clinical assessment alone.3 4