Interpretable Frequency-Aware Deep Learning for Multi-Class EEG Classification of Neurological Disorders

Topic: Brain imaging, connectivity and assessment

Electroencephalography (EEG) records cortical activity with high temporal resolution without radiation and at low cost, which makes it a practical signal for automated screening of neurological disorders. Deep learning classifies EEG with high accuracy, but most models distinguish only two conditions and give no account of their decisions, which limits their use in screening and decision support. This paper presents SpectraXAI, an interpretable frequency-aware deep learning model that assigns resting-state EEG to four diagnostic categories: Alzheimer’s disease, Parkinson’s disease, schizophrenia, and healthy control. The model filters each EEG window into the delta, theta, alpha, beta, and gamma bands and processes every band in a separate convolutional branch with squeeze-and-excitation recalibration. Softmax-weighted band attention and temporal attention then combine the spectral and temporal evidence, and the same attention weights are intrinsic explanations of each prediction. Under a subject-disjoint evaluation protocol that prevents window leakage, the model reached 94.43% accuracy, 96.74% balanced accuracy, and 93.82% macro-F1 on 20,760 windows from 173 unseen subjects while using only 0.10 million parameters. An ablation study confirmed that the attention modules account for most of the improvement over a plain convolutional baseline. Attention weights, feature-attention heatmaps, Grad-CAM, and t-SNE embeddings trace each decision to specific frequency bands, time segments, and learned features, therefore the frequency structure that drives the accuracy also makes the predictions inspectable.

Key findings

• The model filters each EEG window into the delta, theta, alpha, beta, and gamma bands and processes every band in a separate convolutional branch with squeeze-and-excitation recalibration.
• Softmax-weighted band attention and temporal attention then combine the spectral and temporal evidence, and the same attention weights are intrinsic explanations of each prediction.
• Under a subject-disjoint evaluation protocol that prevents window leakage, the model reached 94.43% accuracy, 96.74% balanced accuracy, and 93.82% macro-F1 on 20,760 windows from 173 unseen subjects while using only 0.10 million parameters.
• An ablation study confirmed that the attention modules account for most of the improvement over a plain convolutional baseline.
• Attention weights, feature-attention heatmaps, Grad-CAM, and t-SNE embeddings trace each decision to specific frequency bands, time segments, and learned features, therefore the frequency structure that drives the accuracy also makes the predictions inspectable.

New to this subject? Start with our plain-English guide: Brain scans & tests →

Study details

• Effect size94.43%; 96.74%; 93.82%
• Evidence strengthLow

Read the original

• DOI: 10.20944/preprints202610.1078.v1
• Original source →

Free copy unknown: no free-copy check has an answer for this record yet. That does not mean it is paywalled.

@article{bunterngchit2026-e6cbac35,
  author  = {Bunterngchit, C. and Baniata, L. H. and Baniata, M. H. and Atwan, J. M. and Alali, M. and Alsharaiah, M. A. and Kang, S.},
  title   = {{Interpretable Frequency-Aware Deep Learning for Multi-Class EEG Classification of Neurological Disorders}},
  journal = {Preprints.org},
  year    = {2026},
  doi     = {10.20944/preprints202610.1078.v1},
  url     = {https://doi.org/10.20944/preprints202610.1078.v1}
}

Related studies in Brain scans and imaging

• A Deep Siamese Convolution Neural Network for Multi-Class Classification of Alzheimer Disease Brain sciences · 2020-02-05 · usefulness score 49/100
• Multi-scale 3D CNN with imbalance-aware training for structural MRI-based six-class Alzheimer disease staging Computers in Biology and Medicine · 2026-09-22 · usefulness score 48/100
• Multi-class Alzheimer’s disease (AD) classification using Swin Transformer wavelet and Gray Wolf Optimization (GWO) Intelligence-Based Medicine · 2026-02-26 · usefulness score 40/100
• Multi-modal multi-task learning for joint prediction of multiple regression and classification variables in Alzheimer's disease NeuroImage · 2011-10-04 · usefulness score 36/100
• Deep learning detection of informative features in tau PET for Alzheimer’s disease classification BMC Bioinformatics · 2020-12-01 · usefulness score 57/100
• Small-Data Deep Learning for Alzheimer-Spectrum Classification from Structural MRI: A Feasibility Study Using OASIS Journal of Imaging · 2026-08-03 · usefulness score 42/100

Similar studies

• EEG Pathology Detection via Clinical Concept Bottleneck Models 2026 · Digital health tools
• A Dual-Stream Spatio-Temporal Network with Bidirectional Rectification for Differential Diagnosis of Dementia using EEG 2026 · Brain scans and imaging
• Multimodal deep learning for Alzheimer's disease dementia assessment 2022 · Digital health tools
• Deep learning to detect Alzheimer's disease from neuroimaging: A systematic literature review 2020 · Digital health tools
• Deep Learning in Alzheimer's Disease: Diagnostic Classification and Prognostic Prediction Using Neuroimaging Data 2019 · Digital health tools
• Deep learning-based feature representation for AD/MCI classification 2013 · Digital health tools

More on Brain scans and imaging

• A review that covers this ground Magnetic resonance imaging of iron deposition in neurological disorders Topics in magnetic resonance imaging : TMRI · 2006-02-01 · usefulness score 50/100
• Newest in this subject Large-scale head-to-head comparison of on- and off-target profiles of [18F]Tauvid and [18F]MK6240 tau-PET in aging and Alzheimer’s disease Research Square · 2026-11-01 · Preprint — not yet peer-reviewed · usefulness score 16/100
• Newest in this subject Repetitive negative thinking is associated with amyloid accumulation over 3 years in cognitively healthy adults medRxiv · 2026-10-31 · Preprint — not yet peer-reviewed · usefulness score 16/100

See all studies in Brain scans and imaging →