CTESM

A Hybrid Convolutional-Transformer Approach for Accurate EEG-Based Parkinson’s Disease Detection

This repository contains the Python implementation of the methodology described in our research article on EEG-based classification of Parkinson’s Disease (PD) using a deep learning architecture called the Convolutional Transformer Enhanced Sequential Model (CTESM).

Overview

CTESM integrates convolutional neural networks (CNN), transformer blocks, and long short-term memory (LSTM) layers to capture spatial, temporal, and sequential patterns in EEG data for robust and accurate PD classification.

Repository Structure

The original Jupyter notebook has been modularized into the following standalone Python scripts to enhance readability and reproducibility:

• 1_data_loading_and_preprocessing.py
  Loads EEG datasets and performs preprocessing steps, including normalization and artifact handling.

• 2_feature_extraction.py
  Extracts biologically informed features such as spectral power, band ratios, wavelet coefficients, and statistical measures from EEG signals.

• 3_model_architecture.py
  Defines the CTESM model architecture, integrating CNN layers with transformer-based attention mechanisms and LSTM layers.

• 4_training_loop_and_callbacks.py
  Implements the training loop, loss functions, and callback mechanisms such as early stopping and model checkpointing.

• 5_evaluation_and_metrics.py
  Evaluates the trained model using metrics including accuracy, F1-score, precision, recall, and confusion matrix.

• 6_visualization_and_results.py
  Visualizes training history, plots key EEG features, and generates result summaries for interpretation.

Citation

If you find this work useful, please cite our article:

Bunterngchit, C., Baniata, L. H., Albayati, H., Baniata, M. H., Alharbi, K., Alshammari, F. H., & Kang, S. (2025). A Hybrid Convolutional–Transformer Approach for Accurate Electroencephalography (EEG)-Based Parkinson’s Disease Detection. Bioengineering, 12(6), 583. https://doi.org/10.3390/bioengineering12060583

Datasets

The following publicly available datasets were used in this study:

• Dataset 1: UC San Diego Resting State EEG Data from Patients with Parkinson's Disease from Rockhill et al.
  DOI: 10.18112/openneuro.ds002778.v1.0.5

• Dataset 2: University of Iowa PD EEG Dataset from Anjum et al.
  DOI: 10.1016/j.parkreldis.2020.08.001

---