Connectogram-COH: A Coherence-Based Time-Graph Representation for EEG-Based Alzheimer's Disease Detection

Area of Science:

• Neuroscience
• Biomedical Engineering
• Data Science

Background:

• Alzheimer's disease (AD) is a progressive neurological disorder impacting cognitive function in the elderly.
• Electroencephalography (EEG) signal analysis is a key method for detecting AD by reflecting neural activity.
• EEG signals, being multivariate, are typically treated as multidimensional time series.

Purpose of the Study:

• To propose a novel transformation strategy for EEG signals to improve Alzheimer's disease detection.
• To develop a graph-based, time-resolved representation of EEG data suitable for deep learning.
• To introduce Connectogram-COH as an enhanced coherence-based version of the Connectogram representation.

Main Methods:

• EEG recordings are segmented into time windows and converted into similarity graphs based on signal coherence.
• Adjacency matrices from similarity graphs are flattened into 1-pixel image columns.
• These columns are concatenated to form grayscale images, serving as input for deep learning models.

Main Results:

• The Connectogram-COH representation effectively captures coherence dynamics in multichannel EEG data.
• The method achieves high accuracy in detecting Alzheimer's disease.
• Image-based EEG representations outperform traditional methods in deep learning classification tasks.

Conclusions:

• Connectogram-COH provides a powerful and interpretable method for converting EEG signals into image formats for deep learning.
• The approach enhances Alzheimer's disease detection and shows potential for broader time series classification applications.
• This method offers a promising avenue for EEG-based diagnostics and analysis.