Spectral coherence in normal adults: unrestricted principal components analysis; relation of factors to age, gender, and neuropsychologic data.
This paper demonstrates, by means of Principal Components Analysis (PCA), an objective approach to the reduction of large data sets produced by multichannel spectral coherence analyses. Coherence data, gathered from 371 normal healthy adults using Hjorth/Laplacian referencing during waking eyes-open and eyes-closed states, were analyzed by "unrestricted" PCA where neither spatial nor temporal variance was folded into among subject variance. There was substantial data reduction with our 4416 initial coherence variables for each state reduced to just 150 factors containing approximately 80% of the variance reflecting a 30 fold concentration of information content. Varimax rotation of the first 40 factors, encompassing 50% of the total variance for both states, revealed loading patterns primarily bilateral with no hemispheric bias, relationships primarily between distant single electrode pairs, (although a single electrode to multiple electrode pattern was also observed), and involvement of all spectral bands. Elemental left to right and anterior to posterior coherence patterns, often used on an a priori basis for coherence studies, were not evident among the rotated factor loading patterns. On the basis of high loadings upon extra bipolar artifact channels, 32 factors accounting for approximately 40% of the variance were identified as reflecting artifactual coherence relationships. By multiple regression the 48 non-artifactual factor scores successfully predicted subject age. In general, coherence diminished with age, which may partly explain age-related EEG desynchronization in healthy adults. Coherence factors also predicted 6 of 10 neuropsychologic variables. Gender was successfully predicted by discriminant analysis. No global interpretations about coherence and gender or neuropsychologic function were possible, i.e., almost equal numbers of factors increased as decreased in males as females. PCA derived coherence factor scores are useful for subsequent statistical analyses, but their factor loading plots of cortical coupling may require more experience to fully interpret.