A Minimal EEG Montage for Adult ADHD Detection: Discovering the Frontoparietal Beta–Theta Framework Via Explainable AI

Abstract

Current deep learning models for Attention-Deficit/Hyperactivity Disorder (ADHD) diagnostics rely on opaque, high-density electroencephalography (EEG) arrays, limiting their clinical utility. This study aims to transition adult ADHD screening from an uninterpretable “closed box” into a transparent, scalable framework by identifying the minimal neurobiological features required for accurate detection. We have developed a multimodal deep learning architecture to classify adult ADHD-screened-positive individuals versus neurotypical controls using EEG data recorded during demanding executive functioning tasks. To achieve clinical transparency, we have applied a systematic Explainable Artificial Intelligence (XAI) ablation and occlusion pipeline. This approach has deconstructed the global 26-channel, 5-band network to assess the independent diagnostic sufficiency and contextual necessity of specific spatial lobes and spectral frequencies. Our analysis has revealed that full-scalp, wide-band EEG arrays are computationally redundant for this diagnostic task. The XAI pipeline has reduced the model into a minimal frontoparietal Beta-Theta (FP-BT) framework. We show that classification accuracy comparable to the full-array baseline is maintained using only 16 channels and two localized frequency bands. Moreover, external validation on an independent pediatric EEG dataset confirmed the cross-sample stability of this signature, proving its generalizability across distinct developmental stages and recording hardware. This focused approach is consistent with established neurobiological findings that link executive dysfunction to frontoparietal regions and abnormal brain oscillations. By removing unnecessary electrodes from large-scale EEG arrays, this interpretable method reduces patient preparation time and hardware requirements, making it a practical and biologically meaningful basis for automated ADHD screening tools, pending validation against clinical diagnostic labels.