Multiway array decomposition analysis of EEGs in Alzheimer's disease

Area of Science:

• Neuroscience
• Biomedical Engineering
• Data Science

Background:

• Alzheimer's disease (AD) diagnosis requires robust feature extraction from physiological data, especially for large, diverse populations.
• Standardized diagnostic tools are crucial for multi-site clinical investigations, improving accessibility and mass screening.
• Non-invasive methods combined with generalizable tools can significantly enhance early detection of AD.

Purpose of the Study:

• To apply and evaluate state-of-the-art Multiway Array Decomposition (MAD) methods for feature extraction from electroencephalograms (EEGs) in Alzheimer's disease (AD) patients.
• To compare the efficacy of MAD-extracted features against traditional methods like spectral-spatial average filters (SSFs), AMUSE, and SVD for AD classification.
• To validate the generalizability and diagnostic performance of MAD across multi-site EEG datasets.

Main Methods:

• Two Multiway Array Decomposition (MAD) methods were employed to extract features from multi-site electroencephalograms (EEGs) of Alzheimer's disease (AD) patients.
• Comparative analysis included spectral-spatial average filters (SSFs), algorithm for multiple unknown signal extraction (AMUSE), and singular value decomposition (SVD).
• A feed-forward multilayer perceptron (MLP) was trained for classification and optimization using independent EEG databases.

Main Results:

• Features extracted using MAD demonstrated superior performance compared to SSFs and AMUSE, evidenced by lower root mean squared error (RMSE).
• MAD-based feature extraction achieved up to 100% accuracy in test conditions on an independent EEG dataset.
• The study confirmed the effectiveness of MAD in generalizing across multi-site databases for AD diagnosis.

Conclusions:

• Multiway Array Decomposition (MAD) presents a powerful tool for extracting diagnostic features from multi-site EEGs in Alzheimer's disease (AD).
• MAD offers significant generalization capabilities, crucial for widespread clinical application and mass screening.
• This approach holds promise for discovering novel characterizations of AD and improving diagnostic accuracy.