Accurate deep-learning model to differentiate dementia severity and diagnosis using a portable electroencephalography...

Area of Science:

• Neuroscience
• Medical Technology
• Artificial Intelligence

Background:

• Mild cognitive impairment (MCI) and dementia present significant challenges for aging populations.
• Accessible, cost-effective, and noninvasive diagnostic tools are crucial for early detection.
• Traditional electroencephalography (EEG) systems have limitations in portability and usability.

Purpose of the Study:

• To develop and evaluate a deep-learning-based method for distinguishing healthy volunteers (HVs) from patients with dementia-related conditions using portable EEG data.
• To assess the efficacy of a customized transformer model in classifying cognitive impairment.
• To explore the potential of portable EEG as a practical diagnostic tool.

Main Methods:

• Collected EEG data from 233 participants (119 HVs, 114 patients).
• Transformed EEG signals into frequency-domain features using short-time Fourier transform.
• Trained and evaluated a transformer-based deep learning model using 10-fold cross-validation and a holdout dataset.

Main Results:

• The model achieved an AUC of 0.872 and bACC of 80.8% in cross-validation for HV vs. patient classification.
• Subgroup analyses showed AUCs from 0.812 to 0.898 and bACCs from 74.9% to 86.4%.
• Comparable performance was observed in the holdout dataset, indicating robust generalization.

Conclusions:

• Portable EEG data, when analyzed with deep learning, can effectively differentiate individuals with dementia-related conditions from healthy controls.
• This approach offers a practical and noninvasive method for early detection and classification of dementia.
• The findings support the integration of portable EEG and AI in clinical settings for cognitive health assessment.