Advanced ADHD Detection Using Multivariate Variational Mode Decomposition and Deep Learning: A Novel EEG-Based Framework

This study proposes a novel framework for detecting Attention Deficit Hyperactivity Disorder (ADHD) using electroencephalography (EEG) signals, integrating multivariate variational mode decomposition (MVMD) with machine learning techniques. EEG data, reflecting complex neural dynamics, were analyzed to identify ADHD-specific patterns. MVMD was employed to decompose multi-channel EEG signals from selected channels (C3, C4, P3, P4, T5, T6, O1, O2) into synchronized intrinsic mode functions, capturing inter-channel dependencies. The minimum redundancy maximum relevance (mRMR) algorithm selected non-redundant, discriminative features, enhancing model interpretability. Features were used to train Support Vector Machine (SVM) and Long Short-Term Memory (LSTM) classifiers. Evaluated on a publicly available dataset, the MVMD-LSTM model achieved 95.4% accuracy, 94.8% precision, and 95.1% recall on a holdout test set, outperforming standalone SVM (73.4% accuracy) and LSTM (81.6% accuracy). Compared to traditional univariate decomposition methods, MVMD improved classification by preserving cross-channel neural dynamics. Sensitivity analyses validated the robustness of channel and parameter selections. The selected features, including theta and beta power differences, aligned with known ADHD biomarkers. This framework sets a new benchmark for EEG-based ADHD detection, offering poten-tial for objective, neurophysiologically grounded diagnostics. Future work should validate the approach on diverse cohorts, incorporate resting-state EEG, and explore real-time clinical applications to enhance translational impact, paving the way for reliable diagnostic tools for ADHD and other neuropsychiatric disorders.