Principal component analysis and generalized eigenvector decomposition of the neurophysiology of auditory memory

Magnetoencephalography (MEG) is a neuroimaging technique that records magnetic fields from neuronal activity, offering high temporal resolution for studying brain dynamics. In cognitive neuroscience, event-related designs examine how the brain processes stimuli and performs tasks. MEG is well-suited for these studies, as it captures rapid neural responses to repeated events. However, separating meaningful signals from overlapping neural activity, artifacts, and noise remains challenging. Linear decomposition techniques, such as Principal Component Analysis (PCA) and Generalized Eigenvector Decomposition (GED), offer a data-driven way to extract brain networks. This study applies PCA and GED to MEG data from a melody recognition task to investigate auditory memory. PCA identifies broadband brain networks from data averaged across participants and conditions, with robustness assessed through statistical randomization and single-subject analyses. GED extracts frequency-resolved neural components to examine oscillatory patterns linked to memory and aging. A time-frequency analysis using Morlet wavelets, followed by statistical tests, evaluates differences across conditions and age groups. Results show that PCA reveals distinct brain networks in auditory cortices, the medial cingulate, hippocampus, and prefrontal areas. GED identifies key frequency components related to memory, with age differences in the alpha and beta bands. The consistency of findings supports the validity of these methods for MEG data. This study validates linear decomposition techniques in MEG by comparing the results with existing literature and testing the method's validity under different computational conditions. The approach’s strength lies in its data-driven exploration, without relying on predefined regions of interest, fully utilizing MEG's high temporal and spatial resolution.