Dual-pathway EEG analysis: ML with hybrid DL for adaptive educational interventions
Biomedical Engineering Advances · 2026-10 · 100232
DOI: 10.1016/j.bea.2026.100232 ↗Kurz gefasst
Fachartikel in der Zeitschrift Biomedical Engineering Advances, 2026-10. Verfasst von Carlos Sandoval-Medina und 3 weiteren Autor:innen. Die Arbeit nennt 50 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Diagnosing Attention-Deficit/Hyperactivity Disorder (ADHD) remains a clinical challenge, as does designing appropriate educational interventions for students with this diagnosis. Electroencephalogram (EEG) analysis offers an objective physiological alternative; however, capturing the non-linear, micro-temporal dynamics of attention remains a significant hurdle for current diagnostic models. This study proposes a dual-pathway framework combining a Hybrid Deep Learning (DL) architecture — Convolutional Neural Networks (CNN), a Transformer encoder, and Bidirectional Long Short-Term Memory (BiLSTM) — for binary ADHD diagnosis, with an interpretable Machine Learning (ML) pathway (XGBoost) for micro-temporal attention profiling, applied to 19-channel EEG recordings from 121 children (61 ADHD, 60 controls). The preprocessing pipeline integrated Wavelet Transforms, Independent Component Analysis (ICA), and Common Average Reference (CAR). Data were segmented into five micro-temporal scales (1, 2, 4, 8 and 16s). On the validation fold, the best Hybrid DL checkpoint reached 92.5% accuracy and an AUC-ROC of 0.979; the XGBoost profiling model reached a macro-F1 of 97.2% ( Cohen’s κ = 0.967 ) under subject-wise 5-fold cross-validation for reconstructing five rule-defined attention profiles. A spectral comparison at electrode Cz showed no significant group difference in absolute band power after Bonferroni correction, except for alpha and gamma; this direction was not stable across alternative normalizations. An individual case (subject v263) illustrates how the profiling pathway translates a moderate-confidence ADHD classification (probability 0.85) and a ”Moderate Compensated Fluctuation” dominant profile into a self-regulation-focused intervention plan. By providing interpretable attention maps and performance indices, this system bridges the gap between high-level signal processing and actionable clinical insights, supporting more precise, personalized diagnostic pathways.
Zitieren
APA
Sandoval-Medina, C., Gallegos-Acosta, A., Cardona-Reyes, H., & Parra-González, E. (2026). Dual-pathway EEG analysis: ML with hybrid DL for adaptive educational interventions. Biomedical Engineering Advances, 100232. https://doi.org/10.1016/j.bea.2026.100232
Vancouver
Sandoval-Medina C, Gallegos-Acosta A, Cardona-Reyes H, Parra-González E. Dual-pathway EEG analysis: ML with hybrid DL for adaptive educational interventions. Biomedical Engineering Advances. 2026:100232. doi:10.1016/j.bea.2026.100232
BibTeX
@article{SandovalMedina2026,
author = {Sandoval-Medina, Carlos and Gallegos-Acosta, Alexis and Cardona-Reyes, Héctor and Parra-González, Ezra},
title = {Dual-pathway EEG analysis: ML with hybrid DL for adaptive educational interventions},
journal = {Biomedical Engineering Advances},
year = {2026},
pages = {100232},
doi = {10.1016/j.bea.2026.100232}
}
EUVIMED