NeuroSwin
Swin Transformer–GRU hybrid model for Parkinson's disease classification and brain region localization from EEG signals. (In preparation)
NeuroSwin is an ongoing research project that combines Swin Transformer and GRU (Gated Recurrent Unit) architectures for Parkinson’s disease detection and brain region localization using EEG signals.
Motivation
Parkinson’s disease is a progressive neurodegenerative disorder. EEG provides a non-invasive window into brain activity, but signal analysis requires sophisticated models that can capture both spatial patterns across electrode channels and temporal dynamics across time.
Architecture
- Swin Transformer: Captures hierarchical spatial features from the multichannel EEG data using shifted-window self-attention — particularly effective for structured spatial layouts (electrode maps).
- GRU layers: Model the temporal evolution of neural oscillations, exploiting the sequential nature of EEG recordings.
- Brain region localization: Beyond classification, the model produces attribution maps that highlight which EEG channels (and corresponding brain regions) contribute most to the diagnosis.
Status
In preparation for Biomedical Signal Processing and Control (Elsevier).
Authors: S. B. Shuvo, S. A. Redhila, S. Hossain, A. D. Roy