EEG-Based Dementia Subtype Classification

This repository contains the data and model information necessary to reproduce the results presented in the manuscript "Translational Approach for Dementia Subtype Classification Using EEG Connectome Profile-Based Convolutional Neural Network" published in Scientific Reports. The original dataset before preprocessing can be gound at OpenNeuro ds004504. Here, we are providing pre-processed connectivity maps, model architecture definitions, and trained model weights to facilitate reproducibility.

Contents

• connectivity_maps/: This directory contains the instance-wise connectivity maps used in the study, both with and without outlier rejection. The files are organized by subject and condition (A, C, F).
• model_architecture/: This directory contains the PyTorch definition of the convolutional neural network (CNN) architecture used for dementia subtype classification. The file model.py defines the model structure.
• model_checkpoints/: This directory contains the state dictionary of the best-performing CNN models for each classification task (multiclass, pairwise). The files are named according to the task they were trained for.

Data Description

The connectivity_maps/ directory contains the pre-processed connectivity matrices for each subject and condition. There are two types of data: full/ with all data included and rejected/ with outlier rejected. These matrices were derived from resting-state EEG recordings in the matrix format as demonstrated in the main manuscript. The files are in .mat format. The naming convention is as follows: [condition][subject_id]_I20_[instance_id]

Where:

• condition: The dementia subtype (A, F, C according to AD, FTD, HC).
• subject_id: A unique identifier for each subject.
• instance_id: A unique identifier for each instance from a subject.

Model Description

The model_architecture/model.py file defines the CNN architecture used in this study. The specific hyperparameters used for training are described in the manuscript's Methods section.

Citation

• If you use the original data, please refer to https://doi.org/10.3390/data8060095
• If you use our preprocessed data or models in your research, please cite the manuscript as follows: "Jungrungrueang, T., Chairat, S., Rasitanon, K. et al. Translational approach for dementia subtype classification using convolutional neural network based on EEG connectome dynamics. Sci Rep 15, 17331 (2025). https://doi.org/10.1038/s41598-025-02018-7"

Contact

For any questions or inquiries, please contact the corresponding author as mentioned in the original manuscript.