CNN-based framework for Alzheimer's disease detection from EEG via dynamic mode decomposition.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Alzheimer's disease (AD) and frontotemporal dementia (FTD) are neurodegenerative disorders with distinct EEG changes.
• Eyes-open (EO) EEG analysis is challenging for AD due to spectral instability, unlike FTD.
• Prior research primarily focused on eyes-closed (EC) EEG, yielding higher classification accuracy.

Purpose of the Study:

• To develop an advanced framework for analyzing EO EEG in AD and FTD.
• To improve the classification accuracy of AD and FTD using EO EEG data.
• To investigate the impact of specific frequency bands, like delta, on classification performance.

Main Methods:

• A CNN-based framework incorporating Dynamic Mode Decomposition (DMD) was developed.
• EO EEG data was segmented into shorter temporal windows using DMD.
• A 3D CNN was employed to capture spatio-temporal-spectral representations.

Main Results:

• The proposed DMD-based CNN framework outperformed conventional spectral ML pipelines and FFT-based CNNs.
• Significant improvements were observed in classifying Alzheimer's disease.
• Excluding delta frequencies showed minor gains for AD classification, while FTD classification was unaffected or improved.

Conclusions:

• The DMD-based CNN framework offers a robust method for analyzing EO EEG in neurodegenerative diseases.
• This approach enhances the diagnostic potential of EO EEG, particularly for Alzheimer's disease.
• Delta frequency band dynamics play a differential role in EO EEG classification for AD versus FTD.