Integrated space-frequency-time domain feature extraction for MEG-based Alzheimer's disease classification

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Alzheimer's disease (AD) is a common dementia, often diagnosed using Magnetoencephalography (MEG).
• Previous studies primarily used binary classification and did not fully utilize dual MEG sensor data (magnetometers and gradiometers).
• AD progression involves distinct stages, necessitating multi-class diagnostic approaches beyond simple healthy vs. disease classification.

Purpose of the Study:

• To develop a three-class classification system for Alzheimer's disease (AD), mild cognitive impairment (MCI), and healthy controls (HC) using bimodal MEG data.
• To evaluate novel wavelet-based biomarkers that integrate spatial, frequency, and time-domain information from MEG signals.
• To propose and assess a score-level fusion approach for combining magnetometer and gradiometer data for enhanced diagnostic accuracy.

Main Methods:

• Utilized both magnetometer and gradiometer data from MEG recordings for feature extraction.
• Developed and evaluated a series of wavelet-based biomarkers, focusing on spatial, frequency, and time characteristics.
• Implemented an improved score-level fusion technique to integrate signals from both sensor types for a bimodal recognition system.

Main Results:

• The proposed bimodal system achieved promising three-class classification performance for AD/MCI/HC.
• Gradiometer-derived markers showed superior performance compared to magnetometer-based markers in this preliminary study.
• The left frontal lobe exhibited the highest recognition rate, and spatial markers using wavelet coefficients demonstrated the best performance among the tested markers.

Conclusions:

• The integration of bimodal MEG data and advanced signal processing offers a promising avenue for accurate multi-stage AD classification.
• Wavelet-based spatial markers show significant potential for discriminating between AD, MCI, and HC individuals.
• Further research is warranted to validate and refine this approach for clinical application in early AD detection.