1Universal AI University, Mumbai, India
Email: meenal.pradhan@universalai.in
2Western College of Commerce and Business Management, Mumbai, India
Email: meenal@wccbm.ac.in
3Universal AI University, Mumbai, India
Email: shilpa.joshi@universalai.in
Background: Early identification of mental wellness risks—including cognitive decline trajectories such as mild cognitive impairment (MCI)—is critical for timely intervention. Electroencephalography (EEG) is a scalable, non-invasive modality well-suited to digital neurodiagnostics due to its portability, cost-effectiveness, and sensitivity to neurophysiological changes associated with cognition and affect.
Objective: To evaluate practical EEG-based screening strategies for mental wellness using a reproducible pipeline applied to open-access data, and to provide evidence-informed recommendations for empirical EEG screening workflows.
Methods: The initial sample comprised 88 EEG recordings from the OpenNeuro dataset ds004504. After preprocessing and quality control, 77 recordings were retained, and following further inclusion criteria, the final modelling cohort consisted of 68 subjects. We processed 68 quality-checked EEG recordings using a reproducible workflow. The pipeline included standard preprocessing, extraction of key spectral features (α, β, PAF, TAR), normalization, and checks for feature collinearity. Two baseline models—Logistic Regression and Random Forest—were trained and evaluated with subject-level stratified splits. Model performance was summarized using macro AUC, with sensitivity, specificity, and calibration as secondary measures.
Key Results: The best baseline (Random Forest) reached a macro AUC of 0.5954 on the held out test set (validation 0.7222), with sensitivity 0.2560 and specificity 0.6330 under subject level splits. Feature importance was dominated by spectral markers—particularly the global theta–alpha ratio and central peak alpha frequency.
Conclusion: EEG-based screening using open-access data is feasible but currently yields moderate accuracy. Our best model achieved a macro AUC of 0.5954, with specificity higher than sensitivity, and performance was strongly influenced by preprocessing choices. Spectral markers—especially theta–alpha ratio and peak alpha frequency—were the most informative, underscoring their potential for clinician-centric, explainable neurodiagnostic tools.
Keywords: Electroencephalography (EEG), Digital Neurodiagnostic, Cognitive Health, Mild Cognitive Impairment (MCI), Open Data, Empirical Analysis, Screening Strategies, Machine Learning, Feature Extraction, Cross-Validation.
How to cite this article: Pradhan M, Joshi S. EEG-Based Digital Neurodiagnostics For Cognitive Health: A Reproducible Screening Workflow Using Open Data. Int J Drug Deliv Technol. 2026;16(2s): 1028-1033; DOI: 10.25258/ijddt.16.2s.124
Source of support: None
Conflict of interest: None