TY  - JOUR
AU  - Rehman, H. M. Abdul
AU  - Li, Xiaoli
AU  - Bunterngchit, Chayut
PY  - 2026
DA  - 2026/07/13
TI  - Cross-Frequency Graph-Transformer Networks for Subject-Independent EEG Classification of Neurodegenerative Disorders
JO  - Journal of Artificial Intelligence in Bioinformatics
T2  - Journal of Artificial Intelligence in Bioinformatics
JF  - Journal of Artificial Intelligence in Bioinformatics
VL  - 2
IS  - 2
SP  - 22
EP  - 30
DO  - 10.62762/JAIB.2026.815471
UR  - https://www.icck.org/article/abs/JAIB.2026.815471
KW  - electroencephalography
KW  - neurodegenerative disorders
KW  - graph neural networks
KW  - cross-frequency coupling
KW  - transformers
KW  - subject-independent classification
AB  - Resting-state EEG offers a low-cost, non-invasive biomarker for Alzheimer's and Parkinson's diseases, yet most deep learning models fail to generalize to new patients due to subject-dependent evaluation protocols, isolated frequency-band processing, and the neglect of cross-frequency interactions. We propose CFGT-Net, a Cross-Frequency Graph-Transformer Network. For each EEG epoch, five canonical bands are processed by a graph attention encoder on a learnable phase-lag index adjacency to produce spatial embeddings. A cross-frequency coupling (CFC) attention models band interactions, a temporal transformer tracks their evolution across epochs, and a correlation-alignment loss enforces subject-invariant representations. Evaluated under leave-one-subject-out on two public datasets, CFGT-Net achieves 88.6% accuracy (0.931 AUC) for AD and 85.3% accuracy (0.902 AUC) for PD, outperforming prior methods including DICE-Net. Ablations confirm the adaptive graph, CFC module, and temporal transformer each contribute gains, while learned connectivity highlights fronto-parietal links consistent with known disease patterns. Limitations include small cohorts and residual per-subject variance under LOSO. Modeling adaptive connectivity with cross-frequency coupling improves subject-independent EEG screening and yields interpretable connectivity maps for clinical inspection.
SN  - 3068-7535
PB  - Institute of Central Computation and Knowledge
LA  - English
ER  - 

@article{Rehman2026CrossFrequ,
  author = {H. M. Abdul Rehman and Xiaoli Li and Chayut Bunterngchit},
  title = {Cross-Frequency Graph-Transformer Networks for Subject-Independent EEG Classification of Neurodegenerative Disorders},
  journal = {Journal of Artificial Intelligence in Bioinformatics},
  year = {2026},
  volume = {2},
  number = {2},
  pages = {22-30},
  doi = {10.62762/JAIB.2026.815471},
  url = {https://www.icck.org/article/abs/JAIB.2026.815471},
  abstract = {Resting-state EEG offers a low-cost, non-invasive biomarker for Alzheimer's and Parkinson's diseases, yet most deep learning models fail to generalize to new patients due to subject-dependent evaluation protocols, isolated frequency-band processing, and the neglect of cross-frequency interactions. We propose CFGT-Net, a Cross-Frequency Graph-Transformer Network. For each EEG epoch, five canonical bands are processed by a graph attention encoder on a learnable phase-lag index adjacency to produce spatial embeddings. A cross-frequency coupling (CFC) attention models band interactions, a temporal transformer tracks their evolution across epochs, and a correlation-alignment loss enforces subject-invariant representations. Evaluated under leave-one-subject-out on two public datasets, CFGT-Net achieves 88.6\% accuracy (0.931 AUC) for AD and 85.3\% accuracy (0.902 AUC) for PD, outperforming prior methods including DICE-Net. Ablations confirm the adaptive graph, CFC module, and temporal transformer each contribute gains, while learned connectivity highlights fronto-parietal links consistent with known disease patterns. Limitations include small cohorts and residual per-subject variance under LOSO. Modeling adaptive connectivity with cross-frequency coupling improves subject-independent EEG screening and yields interpretable connectivity maps for clinical inspection.},
  keywords = {electroencephalography, neurodegenerative disorders, graph neural networks, cross-frequency coupling, transformers, subject-independent classification},
  issn = {3068-7535},
  publisher = {Institute of Central Computation and Knowledge}
}