Abstract
Background
Multiple groups have informally observed that the two dominant temporal principal components (PCs) of the pattern reversal VEP obtained with singular value decomposition (SVD) exhibit little inter-subject variability relative to the inter-subject variability of the raw VEP. Using a unique application of SVD we quantify this variability.
Methods
The multifocal VEP stimulus consisted of a dartboard pattern containing 48 checkerboard patches modulated by a 16 bit m-sequence. The checkerboard reversal response for each patch at each of 96 electrodes was calculated by cross-correlation. SVD was applied individually to the data from each of 5 subjects to determine the two dominant PCs per subject. Application of SVD to a matrix formed by concatenating the two dominant temporal PCs of each subject was used to determine ‘composite’ inter-subject PCs.
Results
When SVD was performed on individual subject data within a time window of 220 msec approximately centered on the first peak of the VEP, the dominant pair of temporal PCs accounted for on average 77% of the variance. The composite first and second PCs (as determined via SVD of the matrix containing the pair of dominant PCs from each subject) accounted for 85% of the inter-subject variance.
Discussion
Standard VEPs are notoriously variable from subject to subject. These differences are assumed to primarily be a result of our unique cortical convolution patterns. SVD reveals a robust similarity between early cortical responses.
NIH EY015825, NIH T32 EY07